Introduction
Owing to limited resources, it is important that health care is provided in the most efficient way possible so that the health produced by those services that are provided is maximised. The standard framework by which this is investigated is cost-effectiveness analysis. This examines the costs and effects of different ways of treating patients and summarises this in an incremental cost-effectiveness ratio.
Earlier in the report, we described teams that have been involved in the two studies. Only two were integrated health and social care teams and the others had a broad range of different professional and support staff. The analysis here is restricted to examining the configurations of the teams involved, which may not represent all the different models of IC available.
Different skill mixes across IC teams can potentially have an effect on patient outcomes and costs. Therefore, for example, a greater number of professions and a greater number of professionally qualified staff could improve care by allowing a fuller range of therapies and support to be provided, but it could increase costs because of higher salaries and a greater number of visits.
Background
The simplest approach to cost-effectiveness analysis compares two randomised groups and then calculates the additional costs and additional benefits associated with the most effective treatment for each group. However, this is not possible here as different services have different patient groups. Controlling for these differences, in the presence of patient clustering and cluster-level costs, makes assessment of cost-effectiveness problematic. The analysis plan for this study starts with simple descriptive analysis, then adds in multivariate analyses of costs then ends with an exploratory analysis of cost-effectiveness. Specifically, we sought to produce:
descriptive analysis of mean resource use, costs and quality-adjusted life-years (QALYs) per patient for each team
multivariate analyses examining the relationship between skills mix and patient-level costs
mean costs and QALYs gained will be plotted for each team on the cost-effectiveness plane to highlight which teams are the most cost-effective
multivariate analysis examining the relationship between skills mix and team cost-effectiveness.
Literature review
The COOP study1 identified several important relationships between team characteristics and costs. Using a sample of 1167 patients, it was shown that the average cost per patient increased as the number of different practitioners increased, but fell with a rise in the proportion of qualified staff. Strong relationships were also identified relating to teamworking as measured by the Workforce Dynamics Questionnaire. Exploratory analyses showed no relationship between cost-effectiveness and skills mix variables across the 20 teams studied. However, another study of 403 patients produced different findings, which indicated the opposite effect with respect to the number of different practitioners and no effect relating to qualified staff.1 No assessment of cost-effectiveness was undertaken in the latter study.
Key point 50: the COOP study1 showed that the average cost per patient increased as the number of different practitioners involved in their care increased but, counterintuitively, fell with a rise in the proportion of qualified staff.
The EEICC study2 did not include a formal evaluation of cost-effectiveness but did collect some of the same data as COOP.1 Combining these two data sets is therefore possible and, with the greater number of patient observations and IC teams, it could give us a clearer picture of the relationship between team characteristics, costs and cost-effectiveness.
Secondary analysis of data
Our analysis undertaken within COOP1 and the subsequent study (EEICC)2 had different objectives and collected different sets of data. For example, within COOP1 we collected patient-level data on number and length of all patient contacts broken down by staff professional group. As such, we were able to estimate detailed costs for each patient and explain how these varied by patient and team characteristics. These data were not collected for EEICC,2 which looked more broadly at how overall cost changed with the implementation of an interprofessional management tool.90 Consequently, the COOP1 analysis cannot be directly replicated using data from both studies. The analysis is necessarily restricted to data that were collected in both studies using the same methods.
Resource use and costs
Duration of care data were collected for each patient observed within both studies. The overall staff cost of the service was estimated using reported staffing levels multiplied by unit costs developed from Agenda for Changes scales uprated to include on-costs.91 This was then divided by the annualised total number of patient days recorded in the study periods within each service to calculate a cost per day, then attributed to individual patients based on their individual duration of care within the service.
Outcomes
The EQ-5D was completed at initial assessment and discharge. EQ-5D scores (utilities) were then calculated using the UK tariff based on time-trade-off values. Baseline and discharge utilities were then combined with the length of time between initial assessment and discharge (assuming any change was linear) in order to produce QALYs as well as QALY change relative to baseline.
Cost-effectiveness
Two summary measures of cost-effectiveness were calculated for each team: the average cost-effectiveness ratio and the average net monetary benefit (NMB). The average cost-effectiveness ratio is calculated as the average cost per patient divided by the average QALY gain per patient. The NMB is calculated as the monetary value of health gain produced by the service minus the cost of the service. The monetary value per patient is calculated as the mean QALY gain per patient multiplied by £20,000. The NMB is preferred for explanatory analyses because of the problems with the measurement properties of ratios (i.e. the generation of extremely high ratio values in the presence of small values in the denominator).
It is important to note that the ratios and NMBs reported are not those typically reported in economic evaluations, which are based on incremental values. Incremental values require a counterfactual – what would have happened if an alternative service were provided? – which is not available in an observational study. With incremental figures, we can assess whether or not a particular service is cost-effective (i.e. benefits exceed costs) and those factors that influence cost-effectiveness. With the average figures produced in this study, we can only assess the factors that influence relative cost-effectiveness, we cannot assess whether or not a particular service is cost-effective in absolute terms. For example, using incremental costs and effects, a negative NMB implies that the increase in benefits for a particular service is exceeded by the increase in costs for the service (i.e. it is not cost-effective). In the study here, negative NMBs do not suggest this as we do not know the costs and health effects of alternative courses of action.
Key point 51: the analysis undertaken was to investigate the value of health gain compared with the cost of the service. This cannot demonstrate absolute cost-effectiveness, but allows for comparisons between teams.
Analysis
The initial analysis is descriptive in nature. Mean resource use, costs and QALYs per patient are reported for each team. No hypothesis testing was undertaken for these analyses. Following this, cost functions were estimated to examine the relationship between individual patient costs and skills mix. The dependent variable in these multivariate analyses (cost) will be log-transformed as the data are heavily skewed.
Within these multivariate analyses, skill mix is characterised in terms of the proportion of staff who have a professional qualification and the number of different professions involved in the care of patients. These two explanatory variables attempt to capture notions of substitution and specialisation. These variables are only available at the service level, for example proportion of qualified staff across the service.
Although the preceding analysis examines only costs, cost-effectiveness involves the simultaneous examination of costs and outcomes. A separate set of analyses had been undertaken to look at this issue. Firstly, mean costs and QALYs gained will be plotted for each team on the cost-effectiveness plane to highlight which teams are the most cost-effective. Secondly, a multivariate analysis of cost-effectiveness will be undertaken. The dependent variable will be the NMB of each service, which was described earlier. These analyses should be considered exploratory as they are based on only 33 observations (representing each of the teams) and, therefore, have limited power to identify any relationships. As such, simpler regression models are used which omit several of the explanatory variables used in the patient-level analyses.
Two sets of analyses were undertaken, based on complete cases (i.e. excluding cases with missing data) and an imputed data set (i.e. when missing data are estimated) (see Appendix 2). For the patient-based regression analyses, the team characteristics are incorporated within the multivariate analysis using a random-effects model in Stata.
Results
The teams show a wide variation in average duration of care (ranging from 1 to 138 days) and average cost per patient (ranging from £318 to £11,511) as shown in . All teams saw, on average, an increase in the health of their patients as shown by positive QALY gains.
Summary of resource use, cost and cost-effectiveness for each team
When patient-level costs are examined, clear relationships are seen with respect to HRQoL (as measured by the EQ-5D), impairment and activity (both as measured by the TOM) (). The nature of the relationship is the same for all three variables, i.e. costs initially increase with improving health (or reduced impairment) and then fall for higher levels of health improvement. Considering the staffing variables, there is only good evidence to suggest that increased numbers of different types of practitioners are associated with higher costs.
Regression of patient characteristics and staff input on costs
Key point 52: costs initially increase with improving health (or reducing impairment), then fall for higher levels of health improvement (or impairment). Increased numbers of different types of practitioners are associated with higher costs.
An examination of shows a massive variability in costs and outcomes. The three services depicted by the points joined by the kinked line are relatively more efficient than the others are. Points vertically above the three points defining this efficiency frontier are services that have higher costs for the same outcomes. The frontier represents combinations of the efficient services and, therefore, in theory, all services above the frontier are generating higher costs for patient outcomes than could be produced by the efficient services.
Cost-effectiveness of the different teams. Note: data points represent mean cost and mean QALY gain for each of the teams. The slope of an imaginary line drawn from the origin to the data point equates to cost divided by QALY gain and represents the average (more...)
Key point 53: there is substantial variability in costs and outcomes between the IC teams being studied.
An explanation of this variability in statistical terms was undertaken using the multivariate analysis on average NMB, with the results shown in . These results show no clear explanatory effect of different staffing patterns (i.e. number of practitioners and proportion of skilled staff). However, there is weak evidence of economies of scale, with larger services (proxied by annualised patient days) generating a higher NMB.
Regression of patient characteristics and staff input on NMB
Key point 54: there is no clear effect of different staffing patterns (i.e. number of practitioners and proportion of skilled staff) on NMB. There is weak evidence that larger services generate a higher NMB.
Discussion
The results indicate highly heterogeneous services. Costs per patient range from £318 to £11,511, with patient characteristics being the dominant explanatory effects. Cost-effectiveness is also highly variable with only size of service showing any sign of an explanatory effect. Staffing patterns as described by the proportion of qualified staff and number of different types of staff have little or no influence on costs and relative cost-effectiveness.
These results add to a growing body of literature that show highly variable costs, outcomes and inconsistent relationships relating to staffing patterns.1,87 The massive heterogeneity of services is likely to have an impact on our ability to identify any relationships that exist. Indeed, the variability may also suggest that comparisons between all the services is not sensible and they appear to differ in terms of their purpose. A service with an average duration of care of 1 day cannot be designed to deliver the same care packages as a service with an average duration of care of 138 days. A further qualitative assessment of the different types of teams may be helpful to identify more useful comparisons; however, with only 33 teams, subdividing the sample will make any statistical exploration of differences problematic.
Key point 55: the purposes and composition of IC teams vary substantially and thus it may not be sensible to compare costs indiscriminately between all services. Costs per patient showed great variation and cost-effectiveness was highly variable, with the size of service being the only sign of explanatory effect.
Critique of the methods
There are limitations to the analysis and the comparison with the previous studies. The costs in the COOP1 and this study are based on staff costs only; full budget information was not available for around half of the services. It is possible that non-staff costs vary between services but these variations should not be expected to be related to staffing patterns.
This study did not have patient-level staffing data but, instead, the total staff costs were allocated to patients based on duration of care episode. This inconsistency could lead to different conclusions but this would only relate to patient-level costings; the team-level costings were generated in broadly similar ways. The team-level results are very similar between COOP1 and this study.
The calculation of QALY gains and, hence, monetary benefit assumes a linear change in health between the start and end of the care spell. Both the linear change and the implicit assumption of no health effects beyond discharge are open to question. However, in the absence of good evidence, we feel that further analyses based on alternative assumptions relating to the rate and duration of health changes would be highly speculative. This does, however, add further uncertainties to the results and the conclusions we are capable of making.
One final data problem present in both the EEICC2 and COOP1 study is the potential for missing data. As demonstrated in Chapter 7, there are discrepancies between the estimated throughput of the individual services and the throughput implied by their data returns for our research. These differences suggest either under-reporting within the study or overestimates of patient numbers. These differences could potentially explain some of the large variability in mean cost per patient (which requires an estimate of patient numbers), as seen in Table 1 and .
Conclusion
This study represents the most comprehensive attempt to explain differences in costs and cost-effectiveness across different IC teams. This work, in tandem with other comparable studies, is unable to identify consistent and clear relationships relating to staffing. This suggests that efficiency savings are possible by many services by reducing staff costs to levels seen in comparable teams. However, the identification of comparable teams may be problematic, as there appear to be quite profound differences between them that are not readily explained quantitatively.
Key point 56: further studies of cost-effectiveness of IC services should consider mixed-methods approaches, as the identification of comparable teams might be problematic using quantitative methods alone. We suggest efficiency savings may be possible.
