Abstract
Background
Providing timely, adequate and appropriately-resourced care to people experiencing their first episode of psychosis needs to be informed by evidence-based models of future need in the population. We sought to develop a validated prediction model of need for provision of early intervention in psychosis [EIP] services at the small area level in England up to 2025, based on current epidemiological evidence and demographic projections of the at-risk population.
Methods
We developed a Bayesian population-level prediction tool. First, we obtained small area incidence data on first episode psychoses, aged 16–64 years, from three major empirical studies of psychosis risk (ÆSOP, ELFEP and SEPEA). Second, we identified suitable prior information from the published literature on variation in psychosis risk by age, sex, ethnicity, deprivation and cannabis use. Third, we combined this empirical data with prior beliefs in six Bayesian Poisson regression models to obtain a full characterisation of the underlying uncertainty in the form of suitable posterior distributions for the relative risks for different permutations of covariate data. Fourth, model coefficients were applied to population projections for 2017 to predict the expected incidence of psychotic disorders, aggregated to Commissioning Group [CCG] and national levels. Fifth, we compared these predictions to observed national FEP data from the NHS Mental Health Services Data Set in 2017 to establish the most valid model. Sixth, we used the best-fitting model to predict three nested indicators of need for psychosis care: (i) total annual referrals to early intervention in psychosis [EIP] for “suspected” FEP (ii) total annual cases accepted onto EIP service caseloads, and (iii) total annual new cases of probable FEP in England up until 2025, using small area population projections.
Results
A model with an age-sex interaction, ethnicity, small area-level deprivation, social fragmentation and regional cannabis use provided best internal and apparent validity, predicting 8112 (95% Credible Interval 7623 to 8597) individuals with FEP in England in 2017, compared with 8038 observed cases (difference: n=74; 0.94%). Apparent validity was acceptable at CCG level, and by sex and ethnicity, although we observed greater-than-expected need before 35 years old. Predicted new referrals, caseloads and probable incidences of FEP rose over the forecast period by 6.2% to 25,782, 23,187 and 9,541 new cases in 2025, respectively.
Discussion
Our translational epidemiological tool provides an accurate, validated method to inform planners, commissioners and providers about future population need for psychosis care at different stages of the referral pathway, based on individual and small area level determinants of need. Such tools can be used to underpin evidence-based decision-making in public mental health and resource allocation in mental health systems.
Forecasting of Lung Cancer Incident Cases at the Small-Area Level in Victoria, Australia
