| Donovan, 2020, UK41 | Systematic review | | SR to assess digital interventions designed to help parents decide whether to seek treatment or self-care for children with acute illness | Whether or not digital interventions reduced urgent care service use | Three studies included. For self-triage interventions, sensitivity and specificity of one was 84% and 74% compared with nurse triage and sensitivity was 93.3% and specificity 13% for another when compared with need for ED intervention. No intervention reduced need for urgent care services |
| Aboueid, 2019, Canada42 | Scoping review | General population. Digital platforms using artificial intelligence algorithms for self-diagnosis | To map the scope of digital platforms using artificial intelligence for diagnosis and identify knowledge gaps | Accuracy; regulation; sociological factors; user experience; privacy and security; ethics; design | 19 studies included. Accuracy varied substantially depending on disease and type of platform. Women and more highly educated more likely to choose correct diagnosis from potential options. More likely to be used by those with no access to health care or a stigmatising condition. Regulation lacking |
| Gottliebsen, 2020, Sweden43 | Systematic review | Primary care. Use of digital triage tools by patients accessing primary care | To explore the evidence on features and benefits of patient-operated primary care triage tools | Efficacy; tool type; factors associated with successful implementation; risk factors | 17 studies included. Limited evidence on actual benefits and performance of tools. Available symptom checkers seem to be risk averse and need more real-world testing. Risk of increased or redistribution of workloads. Evaluation guidelines and regulation need development |
| Berry, 2019, USA44 | Retrospective observational | Case records of 168 patients with HIV or hepatitis C attending ED | To test the diagnostic and triage accuracy of five symptom checkers for HIV/hepatitis C | Symptom checker diagnostic accuracy, differential diagnosis and triage advice compared with ED physician diagnosis | All symptom checkers had poor diagnostic accuracy for correct diagnosis listed in top 1, top 3, top 10 or any listing of differential diagnosis (< 20%, < 35%, < 40% and < 45%, respectively) with significant variation between tools. For triage, a significantly higher proportion of hepatitis C found with emergent criteria than HIV (59.7% vs. 35.6%). Symptom checker diagnostic capability inferior to physician diagnosis |
| Yu, 2019, Hong Kong45 | Retrospective observational | 100 random case records for patients attending ED | To test triage accuracy of two online SCs | Sensitivity, sensitivity for emergency cases and specificity of SC compared with ED nurse decision | One SC had overall better sensitivity than the other (74% vs. 50%); sensitivity for emergency conditions was poorer (70% vs. 45%), with low negative predictive value (43% vs. 24%). Current SCs are unsuitable alternatives to ED triage protocols |
| Hill, 2020, Australia46 | Experimental study | 48 medical condition vignettes used in 36 SC tools (1700 diagnosis tests, 688 triage tests) | To determine the quality of diagnostic and triage advice provided by digital SC | Accuracy of diagnosis and triage advice | For 27 diagnostic SCs, correct diagnosis listed first in 36% of tests, in the top three results in 52% and in the top 10 results in 58%. SCs using artificial intelligence algorithms listed the correct diagnosis first in 46% of tests compared with 32% for other types of SC. Rate of first correct result varied from 12% to 61%. 19 triage SCs provided correct advice for 49% of tests. Emergency care (63%; 95% CI 52% to 71%) and urgent care vignette tests provided more appropriate advice (63% and 56%) compared with advice for non-urgent and self care (30% and 40%). Quality of advice provided by SCs is variable; triage advice is risk averse and more likely to recommend higher-level care than needed |
| Berry, 2020, USA47 | Prospective observational | 116 patients attending outpatients clinic with cough | To test diagnostic accuracy of three SC | Diagnostic accuracy of SC alone compared with physician using SC information and three physicians using SC information and clinic notes | For SC only correct diagnosis identified first in list 34.5–16.4% and in top 3 71.6–31%. Mean correct diagnosis first in list 26.4% for SC only, 29.9% for physician using SC information only and 52.6% for physician using SC information and clinic notes. Variation in accuracy between physicians but overall diagnostic accuracy is superior with a combination of SC and physician review |
| Judson, 2020, USA48 | Prospective observational | 950 primary care patients (1159 contacts) in one university health system | To test a rapidly developed online self-triage and scheduling tool for patients with potential COVID-19 | Accuracy of triage; triage time; face-to-face contacts | For symptomatic patients, sensitivity and specificity of detecting need for emergency care was 87.5% and 76.2%, respectively. Sensitivity and specificity of detecting need for self-care was 53.7% and 89.5%, respectively; 61% of asymptomatic patients had no health service contacts within 48 hours. For clinician appointments, time was significantly shorter from start of self-triage to appointment using the online tool than calling a hotline (median 2 minutes vs. 2 hours 15 minutes). Self-triage tools have potential to improve triage efficiency and reduce unnecessary clinic visits |
