Background
Previous modelling work assessing the risks of surgical transmission of CJD was undertaken by ScHARR, culminating in a report in 2006.11 Henceforth, this will be known as the ScHARR report. This report was part of the evidence base appraised by the CJD Advisory Sub-Committee (CJDAS), which produced IPG196.214 This guidance highlighted three high-risk surgical areas: neurosurgery, posterior eye and neuroendoscopy. It was recommended that migration of instruments between sets should be abolished and that single-use instruments were not recommended on the basis of cost-effectiveness with the exception of accessories for neuroendoscopy. A separate recommendation was made that separate sets of instruments should be established for patients born after 1 January 1997 (who are unlikely to have been exposed to the BSE epidemic).
An update of the previous work was undertaken by ScHARR. However, with the agreement of NICE, the current work focuses solely on surgical procedures deemed to be high risk. This update incorporates the latest evidence on key model parameters and assess a range of appropriate strategies and interventions. Reasons for updating IPG196 include the continued evolution of high-quality and less expensive single-use instruments; the lack of adoption of new decontamination methods potentially effective against human prions; the findings of abnormal prion accumulation in the appendixes of patients born after 1996; and anecdotal reports that the recommendations of IPG196 have proved to be difficult to implement, or unachievable, for a number of units. The primary deliverable was a report for a NICE committee that had been convened for the purposes of providing an update to IPG196.
The analyses undertaken assess the potential transmissions of all forms of CJD, which include sCJD, fCJD and iCJD. Throughout the report, any CJD cases that have been caused by surgical transmission will be abbreviated to stCJD.
Elicitation
Many model parameters are subject to considerable uncertainty and were populated following two elicitation sessions undertaken in 2005, one with epidemiological experts and one with decontamination experts. These elicitations were reported in Stevenson et al.11 and the results are repeated in this report. At a meeting of the NICE interventional procedures committee and ScHARR in October 2017, it was decided that the elicitation related to epidemiological parameters should be reconducted to address possible concerns relating to the lack of potential to be misdiagnosed with a different neurodegenerative disease, and with the incubation periods previously elicited. This elicitation session was undertaken on 18 January 2018; the results of the elicitation exercise are in Appendix 4.
Elicitation methods
The 2018 elicitation session was conducted using the Sheffield Elicitation Framework (SHELF). Four experts participated in a face-to-face facilitated workshop. The experts first completed a training exercise (using a quantity known to the facilitator, but unknown to the experts) to familiarise themselves with the elicitation methodology. For each parameter, the experts first recorded their probability judgements individually without conferring. Experts were asked to separately consider lower and upper plausible limits; different scenarios that might lead to high values or low values of the parameter. Probabilities were not attached to these plausible limits; the purpose of eliciting the limits is to mitigate the effects of anchoring and overconfidence, which may occur if a ‘best guess’ is first provided, followed by some assessment of uncertainty around such a guess.
The experts were then asked to provide median values by dividing their plausible ranges into two intervals judged to be equally likely. They were then asked to divide each interval into two further equally likely intervals, hence providing their lower and upper quartiles.
Each expert then declared his/her judgements to the facilitator, who then presented a graphical comparison of all the experts’ individual judgements. Disagreements between the experts’ judgements were highlighted and the experts were invited to justify their own opinions and question each other. Following the discussion, the experts were asked to imagine a rational impartial observer (RIO): an independent observer who has heard and understood the discussion, and on that basis formed his/her own probability judgements. It was for the experts to decide how much weight a RIO would give to the different opinions/arguments that had been stated; if the experts disagreed, with no convincing experts to favour one side over the other, RIO’s uncertainty would be expected to reflect the disagreement.
The experts agreed on a median and quartiles for RIO’s distribution. The facilitator then fitted a probability distribution to these judgements, by choosing a parametric family of distributions and selecting parameter values using a least-squares fit to the cumulative distribution function. The selected distribution was presented to the experts, with feedback in the form of 5th and 95th percentiles (which had not been directly elicited). The experts were asked to comment on whether or not the fitted distribution was an acceptable representation of RIO’s uncertainty and whether or not the level of uncertainty would be justified based on the proceeding discussion. The distribution would be modified as necessary, before being adopted within the ensuing calibration and probabilistic sensitivity analysis (PSA).
Experts were recruited from within the NICE advisory committee. We believed that the workshop format was important to allow for sufficient training of and discussion between the experts. However, owing to the time scale of the project, it was possible to convene only one workshop with four experts.
Cost-effectiveness literature review
The literature searches of bibliographic databases were performed on 14 August 2017 and yielded 1108 citations. Forty-eight citations were obtained for full text retrieval. Evidence from none of the papers was directly used within the model, but some provide context or alternative values and have been detailed in the context of alternative values for the de novo model.
The conceptual model
Previously, authors of this report had undertaken work for the CJDAS to assess the cost-effectiveness of single-use instruments to reduce the risk of vCJD through surgical procedures.11 The paper by Stevenson et al.12 provided further information, where they had utilised a Bayesian approach to take into account data observed since the generation of the results for NICE and submission of the manuscript. As this model was used by the research team in the initial appraisal in 2005, there was a preference to use, or adapt, this model unless it was shown to be not fit for purpose.
Within the literature review by Bennett et al.13 a publication was identified, which was not conceptually different from Stevenson et al.12 but used a system dynamics approach. Bennet et al.13 had three broad aims: (1) to clarify the possible scale of vCJD infection via surgical instruments, (2) to identify the most important factors contributing to this risk and (3) to help prioritise scientific research. Conclusions from the Bennet et al.13 paper were that ‘the risk of surgical transmission of vCJD could not be dismissed’ and that improvements to decontamination ‘should be respectively cost-effective unless vCJD turned out to be a very rare disease’. As the paper by Bennett et al.13 was published earlier than that of Stevenson et al.12 (2005 compared with 2009), this was not preferred to the previous modelling structure.
A further paper by Garske et al.215 was identified that reported that key determinants of future cases of CJD were the number of times an instrument is re-used, the infectivity of contaminated instruments and the effectiveness of decontamination. These results came from a differential equation model that did not consider instrument migration nor the mass transferred to a patient. The former was noted to be a key parameter in Stevenson et al.,12 which also explored uncertainty in the mass transferred and was published later than the Garske et al.215 paper (which was published in 2006) and thus this model was not deemed preferable to that of Stevenson et al.12
Based on the authors’ critique of Bennett et al.13 and Garske et al.,215 there appeared no strong reason to diverge from a model foundation as described by Stevenson et al.12 This model was amended in consultation with the NICE committee, most noticeably to include the possibility that patients may be an stCJD case but could be diagnosed with an alternative neurodegenerative disease.
A schematic of the conceptual model relating to infection transmission in Stevenson et al.12 is shown in ; this model works on an individual patient level for those with CJD infection. The modelling unit was a geographical area representing a population 1/27 of the size of England, which was assumed to have a neurosurgical centre and a posterior eye centre. Population of the model is detailed in Key model parameters. depicts the flows of patients, instrument sets and SIs that have the potential to transmit CJD surgically. Patients have been categorised into three discrete groups: (1) patients who are not infected with CJD; (2) patients who are infected with CJD who are not infectious; and (3) patients who are infected with CJD and are infectious, but asymptomatic. Patients who have clinical CJD would not be operated on with reusable instruments and are assumed to be outside the modelling process. Across time, patients can move (1) from the non-infected state to the infectious but asymptomatic state following an operation with a contaminated instrument, and (2) to the infected and infectious state when the incubation period of the disease for that patient has been reached; additionally, patients can be removed from the model when the CJD infection becomes symptomatic. In all states, patients can die in accordance with background mortality rates applicable to the age of the hypothetical patient. The decontamination cycle removes mass from the instruments and reduces the infectious titre where applicable.
The conceptual model relating to the infection process. Reproduced from Stevenson et al., with permission from Journal of the Operational Research Society.
During the operation, decontamination process and instrument-storing process, instruments may migrate between sets. Furthermore, SIs cannot always be distinguished from similar items in the main instrument set and migration between SIs and instruments from the main set can occur. The rate of instrument migration is important in circumstances where there are multiple contaminated instruments in one set. Therefore, maintaining set fidelity can limit the spread of infection compared with a situation where the contaminated instruments are spread across a number of sets, which can result in a greater number of subsequent transmissions. In order to model this, dynamic SIs were modelled at an individual level, whereas sets were modelled with the possibility of instrument migration.
A key change in the methodology is that where previously patients born after 1996 were excluded from the original ScHARR model, these were explicitly included in the updated modelling work. The rationale for the change was that it may be the case that such patients can be infectious, whereas previously this was not thought possible, and that this explicitly allows an evaluation of the health and cost implications of removing the guidance that patients born after 1996 should use different instrument sets to the remainder of the population.
provides the conceptual model for determining the outcomes for patients who have become infected. There has been a fundamental change in this process since the initial work undertaken by ScHARR, as the possibility that patients who become symptomatic following infection with CJD are misdiagnosed as having a different neurodegenerative disease is included. Further details are provided in following sections.
The conceptual model relating to patient outcome post infection.
The model was run from 1 January 2004, the year at which a proportion of key distributions within the model were elicited, to 2018 in the calibration period. This duration included a 1-year warm-up period from 1 January 2004 to 1 January 2005, which allowed for the possibility that instruments were contaminated with CJD prions at the start of 2005. The expected number of modelled CJD cases (estimated based on the number of transmissions that resulted in clinical infection and the elicited probability of correct diagnosis) between 2005 and 2018 were then compared with those potentially observed in the UK to establish plausible bounds for use within the PSA and then subsequently to determine likelihood ratios for each PSA configuration. This process is described in further detail in a later section (see Appendix 7).
Having established parameter configurations that were plausibly consistent with the number of stCJD cases potentially observed, the model was run for a further 5 years to look at the potential loss of health because of stCJD associated with each strategy evaluated. The 5-year period was agreed with the NICE advisory committee to be an appropriate time period that would be sufficiently long to allow potential cases of stCJD to become apparent, but short enough that the computational time required to generate the results was not excessive and that it did not limit the committee to a decision which could not be changed in the longer term if required. The 5-year period matched the value used by Stevenson et al.12 The measure of benefit was reported in terms of life-years gained and quality-adjusted life-years (QALYs). A lifetime perspective was undertaken for the patients simulated to have high-risk surgery within the 5-year period.
The model was constructed in Simul8 (© 2017 Professional Edition Simul8 Corporation, Glasgow, UK). A NHS and Personal Social Services perspective was taken and both costs and benefits were discounted at 3.5% per annum as recommended by NICE.216
Key model parameters
Parameters relating to the probability and the mass of prions being transferred to surgical instruments
The underlying probability of Creutzfeldt–Jakob disease prions within central nervous tissue in the asymptomatic population
The experts in the elicitation session indicated that the previously elicited distributions relating to the prevalence of CJD prions in all tissue for patients aged 16–39 years in 2005 could still be used for the prevalence of CJD prions in central nervous tissue in 16- to 39-year-olds in the current analysis; although, they acknowledged that the range would produce an overestimate as the probability in all tissue would be greater than that confined to just the CNS. The experts disagreed with the previous experts in whether or not the prevalence would be greatest in the 16–39 years age group compared with the 0–15, 40–69 and ≥ 70 years age groups. The current experts believed that the elicited distribution should be used for all age groups. The distribution used to populate the model is shown in and represents a beta (1.240, 2225.393) for the prevalence. The distribution provides a 95% credible interval (CrI) ranging from 26 to 1875 people per million.
The prevalence of CJD prions within central nervous tissue.
The NICE committee asked for two scenarios to be evaluated, which used different assumptions for the patients born after 1996, henceforth denoted the P96 group. In one scenario, it was assumed that the P96 group were not infectious, as they were assumed unlikely to have been exposed to the BSE epidemic; in an alternative scenario it was assumed that the P96 group had the same probability of being infectious as the general population.
The residual mass per surgical instrument
The ScHARR report11 assumed that the mass on an individual instrument was 2.88 mg of wet-tissue equivalent for instruments used for tonsillectomies, and 1.26 mg of wet-tissue equivalent for instruments used in general surgery. This mass was assumed to be independent of size and complexity. The source for this was ‘provided by Professor Baxter and colleagues from the University of Edinburgh’,11 with these data reported in a different form within Baxter et al.192 It was assumed that the tonsillectomy value was generalisable to the residual mass on a brain and posterior eye surgery. When multiplied by the number of instruments assumed to be in each set, this equated to 51.84 mg of wet-tissue equivalent on brain surgery instrument sets and 25.92 mg of wet-tissue equivalent on posterior eye instrument sets. Each SI would have a wet-mass equivalent of 2.88 mg. These values were assumed fixed.
During discussions on the parameterisation of residual mass, a committee member highlighted a recently published article,196 which suggested that the residual protein mass is likely to be < 5 µg protein mass per instrument side. This is considerably less than that used in the previous ScHARR report,11 which was 576 µg of protein mass (2.88 mg of wet-tissue equivalent).
A preliminary inspection of articles discussing residual mass was undertaken, which indicated that protein mass ranged between 163 and 756 µg (120 instruments) in Baxter et al.192 and between 8 and 91 µg (mean 71.67 µg; 43 instruments) in Murdoch et al.193 Lipscomb et al.195 presented further evidence, which was based on a set from each of nine NHS trusts (260 instruments in total), and reported that 66% of all instruments showed severe contamination in at least one sample area, equating to > 4.4 µg of protein/mm2.
Examining the data in Baxter et al.192 and Murdoch et al.,193 the mean residual protein mass per instrument in 2004 was set to 200 µg (95% CI 150 to 250 µg) in consultation with NICE committee members.
However, the data reported in Smith et al.,196 and further data marked as academic-in-confidence obtained from a NICE committee member (anonymous, May 2018), indicate that there has been a reduction in mass over time for the hospitals where data has been recorded. In discussion with committee members, it was assumed that this change, which is assumed to be related to guidance on keeping instruments moist prior to decontamination, would have occurred in 2012 in line with the purchase of new instruments for those units that had adhered to IPG196. Following discussion with committee members, the mean residual mass for those units that were compliant with guidance to keep instruments moist was assumed to be 10 µg. In the 90% of units that did not adhere to IPG196, it was assumed that two-thirds of these (i.e. 60% of total units) would not keep instruments sufficiently moist and that 200 µg of residual mass would remain on each instrument, with the remaining third (i.e. 30% of total units) adequately keeping instruments moist. It was assumed that the reduction in protein residue on instruments will translate into a reduction in the possibility of transmission of stCJD.
This conceptual model was operationalised by assuming that the mass harvested from a patient from 2012 onwards was 5% (10/200) of the mass assumed to be harvested prior to 2012. Any infectious mass already on an instrument was assumed to remain on the instrument following measures to keep instruments moist.
The proportion of residual mass on brain and posterior instruments that is transferred to a patient
This value was estimated in the original elicitation exercise, which was undertaken to inform the previous work by ScHARR.11 A depiction of the distribution for the proportion of residual mass transferred to the patient is provided in . This has a mean of 31.5% and a 95% CrI 0.4% to 87.1%, showing considerable uncertainty.
The proportion of residual mass transferred to a patient.
The proportion of residual mass on brain and posterior instruments that is removed in a subsequent decontamination cycle
This value was estimated in the original elicitation exercise, which was undertaken to inform the previous work by ScHARR.11 A depiction of the distribution for the proportion of mass transferred to the patient is provided in . This has a mean of 0.9% and a 95% CrI of 0.0% to 4.0%.
The proportion of residual mass removed in a subsequent decontamination cycle.
The proportion of mass on instruments that is replaced with new tissue per brain or posterior eye operation
In accordance with the previous ScHARR model,11 it was assumed that the residual mass on an instrument was in steady state. Therefore, the sum of the mass transferred to the patient and the mass removed in the next decontamination cycle equals the newly acquired mass from the operation. The mean value of the proportion of the mass removed from instruments during an operation is 32.4%, with an estimated 95% CrI of 1.1% to 88.4%. Any SIs that were used were assumed to gather the same mass as instruments in the main set.
Residual mass, proportion transferred to a patient, proportion removed during the operation and the mass harvested during neuroendoscopy
The spreadsheet calculations that were performed to obtain the proportion of mass transferred to the patient and the proportion removed in the next decontamination cycle for rigid neuroendoscopes and flexible neuroendoscopes used in the previous modelling work11 could not be retrieved, but the values used in the PSA were available. These values have been re-used in the modelling, although it appears that there was a discrepancy between the mass transferred from a patient to a rigid neuroendoscope lumen used in the model and the mass that was reported in table 11 of the ScHARR report,11 with the former being 10 times smaller. For the updated work, we have erred on the side of caution and assumed that the greater mass is harvested per operation.
For information, key statistics on the proportion of mass harvested from a patient, the proportion of mass transferred to a patient and the proportion of mass removed in the next decontamination cycle are provided in .
Information relating to the mass transferred to a patient, mass washed off in subsequent decontamination cycles and mass harvested from a patient
Parameters relating to the decontamination of surgical instruments
The assumed infectious titre of tissues containing Creutzfeldt–Jakob disease prions
In the previous modelling undertaken,11 brain and posterior eye tissue were assumed to have 108 ID50 per gram, with this value assumed to be fixed and applied from the moment the patient became infectious to the moment when clinical symptoms of CJD were observed, in which instance reusable instruments would not be used on the patient.
The NICE committee requested, based on the collective experience of its members, that the previous assumptions were amended to allow more heterogeneity in patients who have CJD prions in high-risk tissue. First, the mean infectious titre was varied between 107 and 109 ID50 per gram, assuming a uniform distribution. Second, it was assumed that 20% of patients would have an infectious titre 1 log higher than the mean and that 20% of patients would have an infectious titre 1 log lower than the mean, with the remaining 60% of patients having the mean value. This approach incorporates uncertainty around the mean estimate as well as patient heterogeneity, with individual patient values ranging from 106 to 1010 ID50 per gram.
The effectiveness of current decontamination processes in reducing infectivity
The distributions produced from the elicitation exercise to inform the ScHARR report11 were considered appropriate by the NICE committee. These were split into three categories: (1) the effectiveness of infectivity reduction in the first decontamination cycle, (2) the effectiveness of infectivity reduction in subsequent decontamination cycles and (3) the mass removed in second and subsequent decontamination cycles. The model assumes that there have been no improvements in the reduction in infectivity since 2004.
The effectiveness of infectivity reduction in the first decontamination cycle
The distribution assumed for the infectivity reduction associated with the first cycle of autoclaving is displayed in . This has a mean log-reduction of 2.50 and a 95% CrI log-reduction of 1.42 to 3.58.
The reduction in infectivity in the first autoclaving cycle.
The distribution assumed for the infectivity reduction associated with the first cycle of detergents is displayed in . This has a mean log-reduction of 0.64 and a 95% CrI log-reduction 0.04 to 2.03. Note that detergents used in cleaning neuroendoscopes were assumed to not reduce infectivity.
The reduction in infectivity in the first detergent cycle.
The effectiveness of infectivity reduction in subsequent decontamination cycles
It was assumed in the ScHARR report11 that the second and third autoclaving cycles would reduce prion infectivity, although this would be to a lesser extent than the initial autoclaving cycle. These further autoclaving cycles would occur following a subsequent operation. The log-reduction on the second and third autoclaving cycle was expressed as a proportion of the reduction estimated in the first cycle. The distribution assumed for the multiplier is shown in . This distribution has a mean of 0.157 with a 95% CrI of 0.043 to 0.330.
The proportion of autoclave cycle 1 log-reduction achieved by cycles 2 and 3.
It was assumed in the ScHARR report11 that the second detergent cycle would reduce prion infectivity, although this would be to a lesser extent than the initial autoclaving cycle. The log-reduction on the second and third autoclaving cycle was expressed as a proportion of the reduction estimated in the first cycle. The distribution assumed for the multiplier is shown in . This distribution has a mean of 0.474 with a 95% CrI of 0.047 to 0.931.
The proportion of detergent cycle 1 log-reduction achieved by cycle 2.
The probability of disposing of a reusable instrument
In the ScHARR report,11 it was assumed that following use an instrument had a 1/250 probability of being disposed of (range 1/200–1/300) with all infectious load on the instrument destroyed. In discussions with the committee, it was believed that the serviceable life of a reusable instrument was longer than that previously assumed and the probability of an instrument being disposed of was reduced to 1/2500 with a range of 1/2000–1/3000.
For each instrument that was disposed of in a brain surgery set, it was assumed that between 0% and 12% (sampled from a uniform distribution) of infectious load was removed from the set. For each instrument disposed of in a posterior eye surgery set, it was assumed that between 0% and 25% (sampled from a uniform distribution) of infectious load was removed from the set. The midpoints of these distributions (6.0% and 12.5%, respectively) were chosen such that it was close to the proportion of the set that one instrument comprised. This is (see Parameters relating to instrument migration, costs and safety) a 1/14 probability (7%) for an instrument in a neurosurgery set and a 1/9 probability (11%) for an instrument in a posterior eye surgery set. Uncertainty was incorporated by allowing a range between 0% and approximately twice the midpoint value.
Parameters relating to instrument migration, costs and safety
The instruments assumed on model set-up
In the modelling undertaken for the ScHARR report,11 it was assumed that there were 12 brain surgery sets with 18 instruments assumed to come into contact with potentially infectious mass; 12 posterior eye surgery sets with nine instruments assumed to come into contact with potentially infectious mass; and one rigid neuroendoscope and one flexible neuroendoscope, both of which had a single accessory.
Following discussion with the committee, it was assumed that the number of instruments coming into contact with high-risk tissue in brain operations was lower than previously thought, with the number reduced to 14 instruments (previously 18).
For brain and posterior eye sets, the instrument sets were used in rotation. For neuroendoscopy operations, it was assumed that 75% were undertaken with rigid neuroendoscopes (which can be autoclaved) and 25% were undertaken using flexible neuroendoscopes (which cannot be autoclaved).
Brain and posterior eye sets were also complemented by six types of SI, each of which had six instruments that were used in rotation. During each operation, each SI had a 20% chance of being required.
For neuroendoscopy, IPG196214 recommended that all neuroendoscopy accessories became single use. For simplicity, however, it was assumed that this was not followed, based on the committee’s estimation of units that had adhered to IPG196 and with an assumption that one SI was used in all operations. If a large number of deaths was observed related to neuroendoscopy, this assumption would be amended.
Recommendations on instrument migration and use of supplementary instruments in IPG196
Maintaining the integrity of surgical instrument sets was shown to be a key parameter affecting the incremental cost-effectiveness ratios (ICERs) associated with the introduction of single-use surgical instruments.12 The ScHARR report11 made the following assumptions in relation to set integrity:
That the probability of an instrument being swapped with a similar instrument in a separate set was 50%, while the set was undergoing the decontamination process. This value was selected following discussion with clinicians and review of evidence. When instruments migrate between sets it was assumed that between 0% and 20% (sampled from a uniform distribution) of the infectious material in ‘set A’ would move to ‘set B’, with between 0% and 20% (sampled from a uniform distribution) of the infectious material in set B being moved to set A. These values were chosen as there were approximately 10 instruments in a surgical set, which would be expected to contain 10% of all mass (infectious or not) and that there would be expected uncertainty around the proportion of mass contained on individual instruments.
That when a SI was used, there was a 50% chance that this instrument would join the set with a similar instrument from the set becoming the ‘new’ SI. When this occurs, all infectious load on the SI is added to the set, and between 0% and 10% of the infectious load (sampled from a uniform distribution) in the set is assumed to reside on the new SI. The distribution used to model infectious mass transference as a result of SI migration is associated with smaller mass levels than non-SI instruments.
The model has the facility to alter the levels of set migration following the publication of IPG196,214 which recommended that migration of instruments between sets should be abolished and that SIs that come into contact with high-risk tissues should either be single-use or remain with the set to which they have been introduced. However, owing to logistical and/or financial problems in implementing IPG196,214 these recommendations were not fully adhered to by all hospitals. The model has been set up so that it is assumed that after 2012 no SIs are used for those units that are assumed to adhere to IPG196.
The costs associated with single-use instruments
A NICE committee member stated that the costs of single-use sets are likely to lie in the region of £350–500 and that the cost of a single-use rigid neuroendoscope is £710; no cost was identified for a single-use flexible neuroendoscope (anonymous, May 2018).
The costs associated with reusable instruments
In the ScHARR report,11 it was assumed that a brain surgery set costs £3500 and that a posterior eye set costs £1000. Based on the number of instruments that come into contact with high-risk tissue, the cost of an individual reusable instrument is likely to be in the region of £100–200.
The ScHARR report11 assumed that a reusable rigid neuroendoscope costs £397 and a reusable flexible neuroendoscope costs £9300. More recent prices estimate that a reusable rigid endoscope set including instruments would cost approximately £8850, with a flexible endoscope costing approximately £21,000. Although there will have been inflation during the period, the increase in prices for rigid neuroendoscopes in particular, look high. Clinical advice suggests that in 2005 these were very cheap, disposable rigid neuroendoscopes, but that these have since been withdrawn. This puts downwards pressure on the costs of the reusable rigid neuroendoscopes and, furthermore, it is likely that the volume of sales of neuroendoscopes has decreased resulting in an increase in the price. Whatever the reasons underlying the increase, the NICE committee were comfortable that the prices used in this report was appropriate.
The costs associated with decontaminating reusable instruments
Data provided by a committee member indicated that the cost of decontaminating a reusable instrument was, on average, £0.60 in Scotland (personal communication, May 2018). Assuming that this result is generalisable to England, this would correspond to a decontamination cost of £8.40 for a high-risk tissue brain set and £5.40 for a high-risk tissue posterior eye set.
The costs associated with disposing single-use sets
For simplicity, we have assumed that the costs of disposing of single-use sets are included within the purchase price. Given the relatively wide range in the costs assumed for a single-use high-risk tissue set (£350–500), the authors of this report deemed that this simplifying assumption would not cause significant inaccuracy.
The costs associated with keeping instruments moist
Data reported in Smith et al.196 state that the cost of NHS bags would be £440 per 7355 neurosurgical trays reprocessed, equating to £0.06 per bag. Calculations based on the additional savings that could be made ‘using tap water and tray liner’ suggests that the costs of these elements are also £0.06 per tray. Thus, it has been assumed that the cost of keeping instruments moist was £0.12 per set conditional on using NHS bags, tap water and a tray liner.
The assumed safety of single-use instruments
In the base case it is assumed that the complication rates and outcomes are identical for reusable instruments and single-use instruments. The NICE committee believed this assumption was reasonable.
The costs associated with systems to allow instruments to be tracked
NICE committee members provided data from an unpublished Society for British Neurosurgeons survey and from costs recorded at their own units, which indicated that £750,000 across a 5-year period, including necessary equipment, would be a reasonable estimate (anonymous, May 2018). Sensitivity analyses were intended using £500,000 and £1,000,000.
Parameters relating to the probability of infection, the incubation time and consequences if clinical symptoms appear
The conceptual model of estimating the probability of infection when prions are transferred to the patient
In the earlier ScHARR model,11 the probability of infection was estimated using the mass transferred to the patient (in grams) and the infectious titre of the mass (in terms of ID50 per gram, where an ID50 is the dose required to infect 50% of the susceptible population). It was assumed that the relationship between the number of ID50 and the probability of infection was:
such that 2 ID50 or more would result in a certain infection. In the earlier ScHARR model, the use of a geometric sequence was used, such that 2 ID50 would result in only 75% of patients being infected (1–0.52). However, the committee did not want to use this assumption because the dose to infection was not robustly known and high levels of ID50 transferred could be associated with definite, rather than a high probability of infection, and the committee wished to err on the side of caution. The linear assumption was upheld by the NICE appraisal committee.
The mass assumed to be transferred per operation is detailed in The proportion of residual mass on brain and posterior instruments that is transferred to a patient and the assumed infectious titre per gram is detailed in Parameters relating to the decontamination of surgical instruments.
In a key change from the ScHARR report,11 it was assumed that all patients, regardless of age or genotype, were susceptible to CJD infection.
The incubation period following surgically transmitted Creutzfeldt–Jakob disease infection
The incubation period associated with stCJD was elicited from clinical experts in January 2018 (see Appendix 4). The results are contained in Appendix 1, but are briefly detailed here. The elicited results differed from those previously elicited in that (1) distributions were no longer elicited for each genotype, as it was assumed that a single distribution could cover all genotypes given the incubation period would be affected by the genotype of the recipient, the infecting prion and the infectious dose provided; (2) uncertainty in the mean estimates was formally captured; and (3) it was assumed that all genotypes were susceptible to CJD.
Four incubation intervals were specified; in the base case, each interval was assumed to be equally likely. These were (1) 0.25 to 2 years, (2) 2 to 10 years, (3) 10 to 20 years and (4) 20 to 50 years. Within each time interval a uniform distribution was used on the assumption that each value was equally likely to occur. To allow for uncertainty around the mean incubation period, it was proposed that the first probability of being in the first three intervals would range between 10% and 40%, whereas the probability of being in the fourth interval (20 to 50 years) would lie between 15% and 35%.
As indicated in , should the incubation time be less than the patient’s life expectancy (sourced from the Office for National Statistics217), the patient would display clinical symptoms. Otherwise, the patient would die without CJD symptoms. Each year a proportion of patients incubating CJD die as a result of non-CJD-related reasons, in line with data reported by the Office for National Statistics.217 The probability of non-CJD-related death was dynamic between 2005 and 2014, using the appropriate life table, but was assumed to use life tables from 2014 to 2023.
The infectious period following surgically transmitted Creutzfeldt–Jakob disease infection
The proportion of the incubation period associated with stCJD for which a patient was considered infectious and able to pass CJD prions to instruments was taken from the elicitation session used to inform the earlier ScHARR report.11 This distribution is shown in . The mean of this distribution is 20.0%, indicating that the patient is infectious for only the last 20% of the incubation period. The 95% CrI for this parameter ranged from 15.3% to 25.2%. It has been assumed that the infectious titre of CJD prions is at the maximum value for the entire infectious period.
The proportion of the incubation period during which the patient is infectious.
Estimations of the relative likelihood of returning to high-risk surgery
Patients who are infectious can return to surgery and may do so at a quicker rate than people who have not experienced prior surgery. The earlier ScHARR report11 assumed that (1) people who had previous brain surgery were 43 times more likely to have a further brain operation than people without a history of a brain operation; (2) people who had previous posterior eye surgery were 60 times more likely to have a further posterior eye operation than people without a history of a posterior eye operation; and (3) people who had previous neuroendoscopy were 761 times more likely to have a further neuroendoscopy than people without a history of a neuroendoscopy. These values were based on Hospital Episode Statistics (HES) data that were extracted by a third party (Northgate Information Solutions; Zellis, Hemel Hempstead, UK) and were assumed to be applicable for use in the updated modelling. Having performed sensitivity analyses in the construction of the model, by increasing the relative rates by 10, the model did not appear sensitive to this variable and the values were left at the values used previously.
The assumed costs and quality-adjusted life-years associated with Creutzfeldt–Jakob disease
Once clinical symptoms have developed, it is assumed that patients accrue no further QALYs as a result of the severity of the condition. The earlier ScHARR report11 used a value of £40,000 for the costs associated with treating a case of CJD. This has been updated using the inflationary indices,218,219 which estimate an inflation value of 302.3/240.9 (1.25) between 2005–6 and 2016–17 using the Hospital and Community Health Services index. Data reported in Barnett and McLean220 indicate that costs of additional care and/or equipment were approximately £10,500 per person from invoices received from 33 patients, although the authors of the paper state that ‘local agencies contributions have not been quantified’. This is lower than that assumed in the original ScHARR model, which has been maintained as the base-case value and is favourable to strategies to reduce future stCJD cases. For simplicity, we have assumed that the cost, from a NHS and Personal Social Services perspective, in 2017–18 for a CJD case was £50,000.
The probability that a person with Creutzfeldt–Jakob disease symptoms are not diagnosed with Creutzfeldt–Jakob disease
It is possible that patients with CJD may be diagnosed with another neurodegenerative disease. This possibility was not considered in the initial ScHARR report,11 but was requested following a meeting of the NICE committee. The distribution of patients who were presumed to be diagnosed with another neurodegenerative disease was elicited from experts in January 2018 (see Appendix 1 for full details) for two age bands, with the experts willing to allow the misdiagnosis in the aged 60–80 years category to be the average of the two other age bands: those aged < 60 years and those aged > 80 years. The distribution for those patients aged < 60 years is shown in .
The proportion of patients < 60 years with clinical CJD symptoms who are diagnosed with another neurodegenerative disease.
The mean value is 13.0% with a 95% CrI 0.4% to 26.8%. The distribution for those patients aged > 80 years is shown in . The mean value is 55.0% with a 95% CrI of 18.6% to 88.4%. The simulated distribution for patients aged between 60 and 80 years of age inclusive is shown in . This distribution has a mean of 34.0% with an estimated 95% CrI of 13.5% to 54.3%.
The proportion of patients over 80 years with clinical CJD symptoms who are diagnosed with another neurodegenerative disease.
The simulated proportion of patients aged between 60 and 80 years inclusive with clinical CJD symptoms who are diagnosed with another neurodegenerative disease.
It should be noted that based on the advice of clinical experts on the committee, there has been no change in CJD case ascertainment levels since 2005. This is partially supported by data from the NCJDRSU in the UK (25th Annual Report (see Figure 32), which showed similar age-specific mortality rates between 2005–9 and 2010–16 in those aged 60–64 years and those aged 75–79 years. However, the age-specific mortality rates were higher in the 70–74 years of age group in 2010–16 than in 2005–9, which could be indicative of better ascertainment in recent years. The assumption of equal ascertainment would favour single-use instruments.
In patients who are correctly diagnosed with CJD, the model does not explicitly distinguish between sCJD and stCJD and thus the probability node at the far right of is not contained in the model. However, it is appreciated that patients with stCJD may be categorised as sCJD, and these are used when calibrating the model output to the numbers of observed cases. This is described in more detail in The potentially unobserved number of surgically transmitted Creutzfeldt–Jakob disease cases between 2005 and 2018.
Parameters relating to the numbers of operations that are considered to be high-risk and the characteristics of patients undergoing these operations
The operations considered to be at risk
In consultation with NICE, only high-risk operations are modelled, which have been subdivided into those related to the brain, those related to posterior eye operations and those involving neuroendoscopy. The operations, using HES data to four characters, that are considered to be high-risk were identified by an expert on the NICE committee and are contained in Appendix 5. For brain operations, an expert on the NICE committee grouped the operations into those with normal life expectancy, those where the patient would be expected to survive 18 months, and those with a 50% probability of death at 18 months and a 50% probability of a normal life expectancy.
Only the main procedure codes have been used rather than all the procedure codes, as there is a possibility that more than one high-risk HES code is undertaken within the same operation, using the same instrument set. In the modelling, the HES data have been inflated by 15% as in the ScHARR report11 to take into consideration that not all of the additional operations (between the main procedure and all procedures) are conducted simultaneously with another high-risk code, and also to incorporate operations undertaken by the private sector in non-NHS hospitals.
The estimated number of operations reported within the HES data since 1 January 2005 is provided in . For future years, the average number of operations in the last 3 years was assumed to continue. Operations were assumed to happen at a constant rate throughout the year. It should be noted that the values in are those reported in the HES data as main procedures and have been increased by 15% within the model in line with the earlier modelling undertaken by ScHARR.
The number of operations classified as high risk by the NICE committee (HES data)
Hospital Episode Statistics data provide age breakdowns for each code, with more granularity from the year 2012 than prior to this date. Analysis of these data indicated that the age profile of patients remained relatively stable across time for each of the three brain operation groupings, for neuroendoscopy and for posterior eye operations. Therefore, for simplification, the age profile within 2016–17 was assumed to apply throughout the model. Depictions of each assumed age profile are provided in Appendix 6.
Calibration targets
The observed number of surgically transmitted Creutzfeldt–Jakob disease cases between 2005 and 2018 and the potentially unobserved number of surgically transmitted Creutzfeldt–Jakob disease cases
The observed number of surgically transmitted Creutzfeldt–Jakob disease cases between 2005 and 2018
There are no cases of CJD that have been categorised as stCJD during this period.
The potentially unobserved number of surgically transmitted Creutzfeldt–Jakob disease cases between 2005 and 2018
There are two possible ways in which patients with stCJD can be misdiagnosed. The first is that another neurodegenerative disease is diagnosed; this has been discussed in The costs associated with decontaminating reusable instruments. The second way that stCJD can be misdiagnosed is that a different form of CJD (in particular sCJD) is the presumed diagnosis, as a previous operation may not be recalled. The potential number of patients misdiagnosed as a different form of CJD was investigated.
Data were supplied to a NICE committee member by the NCJDRSU, which detailed whether or not patients who had a diagnosis of CJD since 2005 had a history of neurosurgery or posterior eye surgery, as well as a brief description of the operation. These data were reviewed by a NICE committee member who categorised each patient as having an operation that was of high risk (and, therefore, potentially a stCJD case) or not. The committee member erred on the side of caution, stating whether or not the operation could have the potential to transmit CJD prions to the patient. However, it is possible that some of the cases reviewed occurred in other parts of the UK than England, to which this guidance is limited, which would result in an overestimated calibration target.
For posterior eye surgery, there were potentially 24 individuals who had undergone surgical operations that could have transmitted CJD, although only 10 of these had operations in 2005 or later. The year of the operation is important, as we want to calibrate the model only to cases where the patient had been infected during the modelling period. For brain surgery, there were potentially 13 individuals who had undergone operations that could have transmitted CJD. There were no dates provided for the operations and thus it was assumed that the proportion of operations conducted in 2005 or later that were observed for posterior eye surgery (10/24) were applicable to neurosurgery, which equates to a possible five cases of stCJD since 2005 (rounding to the nearest integer). The sum of these calculations implies that there could have been 15 cases of stCJD transmitted since 2005 that had been misdiagnosed as another form of CJD, or just over one case per year on average.
Categorisation of surgical units, establishing probabilistic sensitivity analysis configurations that are plausible and generating likelihood functions for plausible probabilistic sensitivity analysis configurations
Categorisation of surgical units
Based on the heterogeneity in surgical units adhering to IPG196 and the analyses varying the assumption of whether or not the P96 group (patients born after 1996) could be infectious from birth, six categories of surgical units were defined (denoted S1 to S6). These were:
S1 – a unit adheres to IPG196 and guidance on keeping instruments moist. The P96 group are infectious from birth.
S2 – a unit does not adhere to IPG196 but adheres to guidance on keeping instruments moist. The P96 group are infectious from birth.
S3 – a unit does not adhere to IPG196 nor does it adhere to guidance on keeping instruments moist. The P96 group are infectious from birth.
S4 – a unit adheres to IPG196 and guidance on keeping instruments moist. The P96 group are not infectious from birth.
S5 – a unit does not adhere to IPG196 but adheres to guidance on keeping instruments moist. The P96 group are not infectious from birth.
S6 – a unit does not adhere to IPG196 nor does it adhere to guidance on keeping instruments moist. The P96 group are not infectious from birth.
Based on the opinion of members of the NICE committee it was assumed that, independent of whether or not the P96 group was assumed to be infectious, 10% of units adhered to IPG196 and guidance on keeping instruments moist, 30% of units adhered only to keeping instruments moist and 60% of units neither followed IPG196 nor kept instruments moist. These probabilities were altered in a scenario analysis.
Employing a heuristic to rule out probabilistic sensitivity analysis configurations that would produce implausible results
Owing to the time required for each run [approximately 12 seconds per ‘plausible’ (defined later) PSA configuration] and the number of PSA configurations, random number (RN) streams, scenarios and PSA configurations that would not be compatible with the observed data, heuristics were used to generate the cost-effectiveness results. At all stages, a cautious approach was employed to ensure that potentially appropriate configurations were not prohibited. Appendix 7 describes the methodology using formal mathematical notation, with a lay description provided in the main text.
The initial step was to develop a metric to exclude PSA draws that would clearly be discrepant to the observed data (known cases of CJD that could potentially be attributed to surgical transmission), without having to run these configurations.
Here, a factor to efficiently maximise the likelihood (FML) was established and any PSA configuration with a value greater than the FML value was discarded.
The FML was derived using a combination of parameters related to the infectious titre after a decontamination cycle, the mass transferred to a patient and the prevalence of prion in tissue in asymptomatic patients:
in which
A = mean infectious titre (in log-terms) × log-reduction in infectivity associated with the first autoclaving cycle × log-reduction associated with detergent on the first cycle
B = residual mass on an instrument × (1 – the proportion of residual mass transferred to the patient)
C = the proportion of asymptomatic individuals with CJD prions in their tissue.
In order to generate the FML threshold value, 2000 PSA configurations were drawn from the appropriate distributions and run using 12 RN streams for each of the following scenarios: S1, S2 and S3. Having assessed the likelihood of each of the 2000 PSA configurations producing results consistent with the observed data, it was decided that any draw with a FML value of > e12 would effectively have zero weight and could be discarded without affecting the results. Any draw with a value ≤ e12 could potentially be consistent with the observed data.
Running further analyses to remove probabilistic sensitivity analysis configurations that are potentially consistent with the observed data but generate an implausible number of transmissions when run through the model
In total, 2000 PSA configurations with a FML value of ≤ e12 were sampled. For each configuration, the first RN stream was run, assuming a S3 surgical unit and determining whether or not there was a violation of the permissible limit (VPL) of clinical transmissions for patients aged ≤ 60 years. It was noted that the clinical experts had stated it was implausible that the correct detection rate of CJD was below 50% in this age group and that the assumed maximum number of clinically apparent cases potentially transmitted via surgery, across all ages, was 15. If there was a VPL, the PSA configuration was deemed to be inconsistent with the observed data and the PSA run was discarded. If there was not a VPL, the next RN stream was run with this process repeated until a maximum of 27 RN streams had been run.
The VPL threshold was dynamic and changed as the number of RN streams increased. A large VPL threshold was chosen to reduce the possibility of rejecting viable PSA configurations, while acknowledging that there was also the probability that clinical transmissions had occurred in older patients. The initial threshold for VPL was 36 transmissions, which was constant for the cumulative total across the first six RN streams. From RN streams 7 to 13, the VPL threshold was increased to 40; from RN streams 14 to 17, the VPL threshold was increased to 45; from RN streams 18 to 23, the VPL threshold was increased to 55; and for RN streams 24 to 27 the VPL threshold was increased to 66. This resulted in 509 out of the 2000 PSA runs that all had an FML ≤ e12 being potentially consistent with the observed data. These are denoted ‘plausible’ PSA configurations.
Calculating the likelihood of each plausible probabilistic sensitivity analysis configuration being consistent with the observed data
Approximate Bayesian computation methods were used to estimate the likelihood of a PSA configuration being consistent with the observed data. Full details are provided in Appendix 7. A likelihood ranges from 1, where the simulated number of transmissions that are clinically detected are entirely consistent with the number of observed cases, to zero where the simulated number of transmissions that are clinically detected cannot be consistent with the number of observed cases. Within this decision problem, any PSA configuration that produces ≤ 15 transmissions that result in clinical symptoms would have a likelihood of 1, whereas any PSA configuration that produced > 30 transmissions that result in clinical symptoms, in patients than < 60 years of age, would have a likelihood of zero.
The likelihoods for each PSA configuration are shown in . These have been ranked in descending order and have been curtailed at 250 of the 509 PSA configurations. A large proportion of the PSA configurations that were not rejected have likelihoods close to zero, which offers support to the belief that it was unlikely that potentially appropriate PSA configurations were discarded. For information, the lowest likelihood was 10–12 where the P96 group was assumed to be infectious and 10–13 where the P96 group was assumed not to be infectious.
The likelihoods of the PSA configurations being compatible with the observed data (curves are drawn on top of each other).
Generating estimates of the expected numbers of future surgically transmitted Creutzfeldt–Jakob disease, life-years lost and quality-adjusted life-years lost
The likelihoods associated with each PSA sample were multiplied by the results (future stCJD deaths, life-years lost and QALYs lost) produced when using that PSA sample and these were added together and divided by the sum of the likelihood to produce expectations for the combined results.
Exploring the uncertainty in the results produced within the base-case analyses
In order to explore more pessimistic scenarios, the maximum value across all of the 509 PSA configurations of the number of QALYs simulated to be lost multiplied by the likelihood of the PSA was also calculated. These values are necessarily greater than the expectations, which use the average value multiplied by the likelihood of the PSA rather than the maximum value. Generating CIs around the mean of each output was more complex owing to the use of likelihoods, as not all of the 509 scenarios were weighted equally. In order to provide an indication of the width of the CI (which would need to be halved if only looking at increasing or decreasing the value from the mean), an approximation was made, which is detailed in Appendix 7, that involved simulation to translate each PSA likelihood into either zero or 1 and then using statistical techniques to estimate a CI.
Exploring the probability that each type of surgical unit was the most cost-effective
Exploratory analyses were undertaken to provide indicative probabilities that each type of surgical unit (one of S1, S2 and S3, or S4, S5 and S6) or moving to single-use instruments were most cost-effective across a range of cost-per-QALY thresholds. This analysis assumed that a surgical centre was a S3 (S6), meaning that expenditure was required to move to S1 or S2 (S4 or S5). The probabilities were calculated assuming that the weight applied to each of the 509 PSA values would be provided to the surgical unit or single-use instrument scenario that was most cost-effective at a chosen cost-per-QALY threshold. The summated total of weights for each option was divided by the sum of the total weights to provide a probability of being most cost-effective, which summate to 1.
Exploring the changes in the results produced with alternative assumptions relating to the assumed distribution of surgical units between the assumed decontamination levels
In the base-case analyses, it was assumed that 10% of surgical units would both follow IPG196 and keep instruments moist; 30% would not follow IPG196; and 60% of surgical units neither kept instruments moist nor followed IPG196. The NICE committee requested that a scenario analysis be run that changed these proportions to 50%; 30%; and 20%, respectively. Thus, in this scenario analysis half of surgical units both followed IPG196 and kept instruments moist.
Strategies modelled
In consultation with the NICE committee, the following strategies were run:
Do nothing, assuming that the current situation is maintained with respect to surgical centres’ adherence to IPG196.
Full adherence to IPG196, and guidance on keeping instruments moist for those units where this is not followed.
Full adherence to keeping instruments moist for those units where this is not followed.
Removal of the requirements to have separate instrument sets for the P96 group.
Modelling interventions that prohibit the possibility of stCJD. These are likely to take the form of the introduction of single-use instruments or the introduction of a decontamination product during the sterilisation process that is completely effective.
Within this report ‘adherence to IPG196’ is a slight misnomer, as the modelled scenario does not assume that neuroendoscopy instruments are single-use. However, for brevity, we have used the term ‘adherence to IPG196’.
Based on advice provided by the NICE committee, it was assumed that the quality of single-use and reusable instruments were equivalent.
Based on advice provided by the NICE committee, no modelling of decontamination products was conducted other than that contained in strategy five. The reasons for this were multiple. First, there was a lack of homogeneity in the identified studies of decontamination products in terms of prion strains, drying times, infectious titre of the material used, time and temperature of the exposure to the decontaminant, dose of the decontaminant, observation period, substrate used, assay and infectivity detection method used. Second, the findings for some agents inevitably differed both within and between studies, owing to the described heterogeneity (see results for Rely+On depending on the assay or NaOH depending on the prion strain). Identifying the most ‘efficacious’ decontaminant, requiring comparison across agents, was, therefore, not possible. Third, as far as we could tell, the majority of the decontaminants (and combinations thereof) were not commercially available but had been developed for the laboratory tests, whereas others that did exist as distinct products were few and, in some cases, were no longer on the market, for example Rely+On. Fourth, uptake of additional decontaminant solutions might be very low in practice owing to requiring an extra step in the sterilisation process. Therefore, major concerns affected the certainty and generalisability of the evidence on decontaminants for reducing the risk of prions in surgery. The authors wanted to provide an indication of the potential prices that could be cost-effective if a completely effective decontamination product was commercially available and so explored this in strategy 5.
Epidemiological results
For each PSA scenario the number of transmissions by age group that resulted in clinical symptoms (whether correctly diagnosed as CJD or not), the number of life-years lost and the number of discounted QALYs lost were simulated through the mathematical model. These results were then weighted by the likelihood, with the sum of these values divided by the sum of the likelihoods.
The epidemiological results presented are based on an individual surgical unit. Units are denoted S1 to S6 (defined in Categorisation of surgical units) to represent the combinations of the unit’s adherence to IPG196; whether or not instruments are kept moist; and whether or not it is assumed that the P96 group is infectious. It has been assumed that there are 27 units in England.
It is assumed that the answers produced will contain Monte Carlo sampling errors and that further RN streams and PSA configurations would provide more accurate answers. However, we believe that the results presented are sufficiently robust to draw conclusions. The base-case results assume that there may have been up to 15 deaths attributable to stCJD between 2005 and 2018.
Base-case results
The base-case results are provided in and relate to the period 2019–23, as agreed with the NICE committee. The estimated values are presented in columns two to four; these are calculated using all PSA configurations (n = 509) and all RN streams (n = 27). The values of simulated deaths as a result of CJD infection, which were weighted by their likelihood that the transmissions of CJD modelled between 2005 and 2018, matched the observed data. The final column contains a value that represents the maximum value across the PSA configurations of the simulated deaths in that PSA multiplied by the likelihood of that PSA. Note that the maximum deaths across the P96 and the non-P96 group may not equal the maximum values for both the P96 group and the non-P96 group individually. The values are per surgical unit and need to be multiplied by 27 to represent values for England.
Base-case results per surgical unit
Interpretation of the base-case results
As anticipated, fewer deaths as a result of stCJD were estimated when IPG196 was followed and when residual mass was reduced. Thus, in terms of future deaths as a result of stCJD, S1 had fewer deaths than S2, which had fewer deaths than S3, and S4 had fewer deaths than S5, which had fewer deaths than S6. Furthermore, as anticipated, when the P96 group was assumed not to be infectious there were fewer projected deaths as a result of stCJD; that is, S1 had more deaths than S4, S2 had more deaths than S5 and S3 had more deaths than S6.
Those units that followed IPG196 and kept instruments moist (S1 and S4) had 0.052 and 0.038 future deaths caused by stCJD, respectively. Where IPG196 was not followed but instruments were kept moist, there was an increase in future deaths as a result of stCJD of 0.035 when the P96 group was deemed infectious from birth and 0.040 when the P96 group was not deemed infectious from birth. Assuming IPG196 was not followed, failure to keep instruments moist was associated with an increase in the estimated numbers of future deaths compared with not following IPG196 increased by 0.343 when the P96 group was deemed infectious from birth and 0.310 when the P96 group was not deemed infectious from birth. From these results, it is apparent that ensuring that instruments are kept moist has a large impact on the risk of future transmissions.
It is of note that the number of potential stCJD infections in the P96 group is not necessarily zero, even when these patients are assumed not to be infectious. This can occur when a P96 patient is infected via an operation prior to 2012, the date at which the new instrument sets for the P96 patients were introduced. Such a patient could then have a further high-risk operation while in the subclinical but infectious period, which could have infected P96 patients.
The circumstances in which the maximum future deaths predicted within the model were explored. A high number of future deaths were associated with the prevalence of CJD prions in their tissue being very low: < 1 per 200,000 people had prions in their tissue. In these PSA runs, no infectious people had entered the system between 2004 and 2018; this resulted in no infections and thus these PSA runs have a likelihood of 1 of matching the observed data. In the 2019–23 period, infectious people were simulated to have an operation in some RN streams, which resulted in infections and deaths. The number of deaths was greater where IPG196 was not followed and where instruments were not kept moist. The maximum number of future deaths multiplied by the likelihood is expected to be associated with approximately 10 times more deaths than the expectation. For completeness, the best-case scenario would be that there were no further deaths, which applies for all types of surgical unit.
Uncertainty in the mean number of QALYs gained was explored as described in Exploring the uncertainty in the results produced within the base-case analyses and Appendix 7. The width of the CI around the mean estimate of QALY loss was estimated to be 0.25 for S1 units, 0.58 for S2 units, 2.07 for S3 units, 0.19 for S4 units, 0.58 for S5 units and 1.89 for S6 units. To explore the relationship between the number of PSA samples and the width of the CI, a randomly selected PSA was removed with the remaining 508 split into two groups of 254. The widths of the CIs for each of the two groups were 0.32 and 0.40 for S1; 0.87 and 0.78 for S2; 3.02 and 2.85 for S3; 0.25 and 0.29 for S4; 0.78 and 0.87 for S5; and 2.76 to 2.62 for S6. This indicated that approximately doubling the number of PSA configurations had led to a reduction in the width of the CIs by approximately 30%. The CIs produced from the 509 PSA configurations were not believed by the authors of this report to be large enough to endanger the conclusions of the analyses are endangered. Given this, it was believed that further reductions in the width of the CIs through running further PSAs were not required.
Scenario analyses using the base case as the foundation
Eight scenario analyses were run, with the change within a unit being assumed to happen instantly at midnight on the 31st December 2018. These scenarios comprised strategies to follow IPG196 and/or reduce the residual mass on instruments, and estimated the effect of removing the guidance on having different instrument sets for the P96 group from the remaining patients. The results of the scenario analyses are presented in . The results are presented in terms of surgical centres; these values would be needed to be multiplied by 27 in order to form estimates for England.
Results of the scenario analyses per surgical unit using the base case as the foundation
Interpretation of the scenario analyses results using the base case as the foundation
These results are subject to Monte Carlo sampling error, particularly in relation to the RNs exhausted within a simulation. For example, in the scenario analysis that changed a unit from S2 to S1, at the start of 2019 this model run will have used significantly more RNs than a comparison with S1 alone. This is a result of the RNs required in selecting from 2012 onwards, the SIs used in an operation and the migration of instruments between sets (which is a feature of S2 but not of S1). This misalignment of RNs between runs will result in different simulated outcomes.
Despite the presence of Monte Carlo sampling error, the results generated are broadly consistent between comparable units, which offers support that the values are relatively robust. However, caution is advised in trying to interpret differences in the results of the scenario analyses (see ) and the base-case results (see ), as these differences could be artefacts of the RNs selected. Significantly more computational time would be required to provide an accurate comparison of the scenario analyses and the base-case results; this was beyond the time scales of the project.
Scenario analyses using an alternative distribution of surgical unit compliance with following IPG196 and keeping instruments moist
As described in Exploring the probability that each type of surgical unit was the most cost-effective, the distribution that was assumed in relation to following IPG196 and guidance on keeping instruments moist was changed to provide an indication of the sensitivity of the epidemiological results to these parameters. The results for the expected number of QALYs lost as a result infections occurring between 2019 and 2023 are shown for the base-case and the alternative scenario in . The results are very similar, as will be the costs associated with each strategy and, as such, no analyses of the alternative scenario will be provided as these are highly comparable to those of the base case.
Comparing the QALYs lost within the base case and when using an alternative assumption related to the distribution of surgical units following IPG196 and in keeping instruments moist. The percentages in the figure key refer to the proportion of units (more...)
Cost-effectiveness results
The presented results have been grouped by type of surgical unit (from S1 to S6). Within each category, evaluated strategies are compared incrementally if appropriate. In addition to the base-case results, sensitivity analyses have been run that change the values of parameters and threshold analyses have been performed to determine at what price for a single-use kit, or cleaning solution that was 100% effective, the cost per QALY gained would equal the chosen cost-effectiveness threshold.
In all analyses, the additional costs have been calculated considering the following elements: the costs of single-use sets; the disposal costs of reusable instruments; the costs of autoclaving reusable instruments; and the costs associated with symptomatic stCJD.
When cost-per-QALY values have been calculated, these are compared with threshold values used within common NICE evaluations. These are £30,000 within a standard technology appraisal, although this can potentially be raised to approximately £50,000 if the end-of-life criteria are met,216 and between £100,000 and £300,000 for highly specialised technologies.221
Parameter values within the base-case cost-effectiveness results
The parameter values used within the base-case estimate of the cost-effectiveness of various strategies are shown in . It is noted that the number of operations were discounted such that sensitivity analyses on the values could be performed without re-running the model. On completion of the runs, it was discovered that the number of instruments disposed of within the run was not saved to file. As such, an estimate of this was calculated rather than being directly taken from the model; it is unlikely that this limitation will influence the results owing to the relatively small values involved.
Parameter values used within the cost-effectiveness analyses
The base-case cost-effectiveness of strategies for reducing the likelihood of surgically transmitted Creutzfeldt–Jakob disease
Results for S1 and S4 units
For surgical units that adhere to IPG196 and keep instruments moist, the only strategy currently available to reduce the potential for stCJD is to use single-use instruments. Based on the values reported in , it is estimated for a S1 unit that the additional net cost of single-use instruments per unit would be £1,814,139, which would produce an expected 0.459 QALYs, thereby resulting in a cost per QALY gained of £4.0M. For a S4 unit, the net cost was similar (£1,814,545) with fewer QALYs gained (0.275), resulting in a cost per QALY gained of £6.7M. Both cost-per-QALY estimates are markedly higher than the thresholds commonly used by NICE.
Results for S2 and S5 units
For surgical units that do not adhere to IPG196 but keep instruments moist, two strategies are currently available to reduce the potential for stCJD: the use of single-use instruments and adhering to IPG196.
Based on the values reported in , it is estimated for a S2 unit that the additional net cost of single-use instruments per unit would be £2,562,829, which would produce an expected 0.874 QALYs, thereby resulting in a cost per QALY gained of £2,933,530. For a S5 unit, the net costs were similar (£2,563,238) with fewer QALYs gained (0.736), resulting in a cost per QALY gained of £3,484,476. Both cost-per-QALY estimates are markedly higher than the thresholds commonly used by NICE.
For a S2 unit, adherence to IPG196 is estimated to have a net cost of approximately £750,000 and provide an increase in QALYs of 0.415, resulting in a cost per QALY of approximately £1.8M. For a S5 unit, adherence to IPG196 is estimated to have a net cost of approximately £750,000, an increase in QALYs of 0.461, resulting in a cost per QALY gained in the region of £1.6M. Both cost-per-QALY estimates are markedly higher than the thresholds commonly used by NICE.
Results for S3 and S6 units
For surgical units that are neither adhering to IPG196 nor keeping instruments moist, three strategies are currently available to reduce the potential for stCJD: the use of single-use instruments; adhering to IPG196 and keeping instruments moist; and keeping instruments moist.
Based on the values reported in , it is estimated for a S3 unit that the additional costs of single-use instruments per unit would be £2,550,760, which would produce an expected 4.009 QALYs, thereby resulting in a cost per QALY gained of £636,292. For a S6 unit the costs were similar (£2,552,043) with fewer QALYs gained (3.485), resulting in a cost per QALY of £732,364. Both cost-per-QALY estimates are markedly higher than the thresholds commonly used by NICE.
For a S3 unit, keeping instruments moist is estimated to produce a cost saving (as the costs of potential prevented CJD cases outweighed those associated with keeping the instruments moist) and to provide an increase of 3.135 QALYs, suggesting that keeping instruments moist is dominant (lower costs and more QALYs produced). For a S6 unit, there was also an expected cost saving of an increase in QALYs of 2.749, resulting in keeping instruments moist being dominant.
For a S3 unit, having initially kept instruments moist, the cost-effectiveness of adhering to IPG196 would be similar to that of moving from S2 to S1, that is in the region of £1.8M per QALY gained. For a S6 unit having moved to a S5, the cost per QALY gained of adhering to IPG196 would be in the region of £1.6M.
Estimating the probabilities that each type of surgical unit or using single-use instruments are the most cost-effective strategies assuming that a centre does not currently follow IPG196 nor keep instruments moist
The probabilities of each surgical unit and using single-use instruments being the most cost-effective are provided in when it is assumed that the P96 group are infectious, and in when it is assumed the P96 group are not infectious. These results assume that all surgical units are currently S3 or S6. Both figures have similar characteristics in that S2/S5 (units that keep instruments moist but do not follow IPG196) have the highest probability of being cost-effective, followed by units that continue to ignore IPG196 and those that do not keep instruments moist. Even at high cost-per-QALY thresholds, the probability that single-use instruments are the most cost-effective is negligible. The probability of being most cost-effective accord with the scenarios (S2 and S5) that are estimated to be the most cost-effective.
The probabilities that S1, S2, S3 and single-use instruments are the most cost-effective at a range of cost-per-QALY thresholds. SUI, single-use instrument.
The probabilities that S4, S5, S6 and single-use instruments are the most cost-effective at a range of cost-per-QALY thresholds. SUI, single-use instrument.
Sensitivity analyses performed on the base-case results
Having observed the ICERs that were presented in terms of cost per QALY gained produced in the base case, the sensitivity analyses that was performed explored a combination of all of the values that were more favourable to single-use instruments. Thus, the cost of a CJD case was increased to £70,000; the average cost of a reusable instrument was assumed to be £200; and the cost of a single-use set was assumed to be £350. Note that these sensitivity analyses change the costs only and that the benefits in QALYs are assumed to be constant.
Sensitivity analyses results for S1 and S4 units
The ICER for single-use instruments for a S1 unit became £2.9M, whereas the ICER for a S4 unit became £4.9M. Neither value was below the commonly used NICE thresholds.
Sensitivity analyses results for S2 and S5 units
The ICER for single-use instruments for a S2 unit became £2.4M, whereas the ICER for a S5 unit became £2.9M. Neither value was below the commonly used NICE thresholds.
For a S2 unit, adherence to IPG196 is estimated to have an ICER of approximately £1.2M, whereas for a S5 unit this ICER was approximately £1.1M. Neither value was below the commonly used NICE thresholds.
Sensitivity analyses results for S3 and S6 units
For a S3 unit, keeping instruments moist remains a dominant strategy. This is also the case for a S6 unit.
For a S3 unit, having initially kept instruments moist, the cost-effectiveness of adhering to IPG196 would be similar to that of moving from S2 to S1, that is in the region of £1.2M per QALY gained. For a S6 unit, the cost-per-QALY of adhering to IPG196 would be in the region of £1.1M, which is similar to moving from a S5 to a S4 unit. These values are similar rather than identical, as there may be more infectious material on instruments in the S3 and S6 units than in S2 and S5 units.
Threshold analyses on the costs of single-use sets or a completely effective cleaning solution
Analyses were performed to indicate the cost at which a single-use set (including disposal costs) would be cost-effective at cost-per-QALY thresholds of £30,000, £50,000, £100,000 and £300,000. These results are identical to the threshold cost of a cleaning solution that was 100% effective at removing CJD prions, as both approaches (single-use instruments and the cleaning solution) are assumed to prohibit CJD infection via surgery.
The results are presented for each unit type, by four cost-per-QALY thresholds, and for the base case and for a scenario analysis that was more favourable to reusable instruments and a completely effective cleaning solution. The results are presented in . Caution must be used in interpreting these results, as options other than single-use instruments or a completely effective cleaning solution exist. For example, moving from a S6 to a S5 (or S3 to a S2) is estimated to be a dominant strategy and thus the thresholds for single-use sets for a S6 unit or a S3 unit are redundant, although these have been presented for information.
Threshold analyses on the cost of single-use sets (including disposal costs) and a completely effective cleaning solution
It is seen that in units where instruments were kept moist, a single-use set price would need to be in the region of £50 to have a cost per QALY below £300,000; to be below a cost per QALY of £30,000, the cost of a single-use kit would need to be in the region of £10.
Threshold analyses on the costs of adhering to IPG196
For S2 and S5 units, analyses were performed to indicate the cost at which adhering to IPG196 would produce an ICER equal to a chosen threshold. These results are presented for the S2 and S5 unit types, by four cost-per-QALY thresholds, and for the base case and for the scenario more favourable to reusable instruments and a completely effective cleaning solution. The results are presented in . The estimated cost of implementing IPG196 is estimated to be £750,000, which is greater than the threshold values provided in .
Threshold analyses on the cost of implementing IPG196
Estimating the cost-effectiveness of removing the need for the P96 group to be operated on with separate instrument sets
The data reported in and indicate that there would be fewer deaths and marginally more QALYs lost when the recommendation that the P96 group are operated on using different instrument sets is removed and the P96 group is considered infectious on model entry. These results lack face validity, particularly in relation to QALYs lost as younger patients can lose more QALYs, and is caused by Monte Carlo sampling error as a result of the misalignment of RNs. Conversely, where the requirement for different instrument sets is removed, given the assumption that the P96 group are not infectious on model entry there are an additional 0.18 QALYs lost although marginally fewer deaths.
The computational time required to provide an accurate estimate for both the number of deaths (which may be equal in both scenarios) and the QALYs lost is far beyond the resources assigned to this work. As such, the results should be interpreted with caution, although currently there is no indication that removing the recommendation related to separate instrument sets would greatly influence the numbers of predicted CJD cases.
