Improving psychosis prediction using primary care consultation data (MAPPED Study)

Problem

Clinical and social outcomes of psychosis are often poor, and many risk factors are difficult to modify. Duration of untreated psychosis (DUP) has been linked to poorer prognosis and is potentially modifiable. Speedier referral from primary care to secondary mental health services is an important component of DUP. We previously found that specific prodromal characteristics were strongly associated with a later diagnosis of psychosis; attention deficit hyperactivity disorder-like problems, bizarre behaviour, blunted affect, depressive-like problems, role functioning problems, social isolation, mania, obsessive compulsive disorder-like problems, disordered personal hygiene, sleep disturbance and suicidal behaviour (including self-harm), cannabis use and cigarette smoking. The positive predictive value of these characteristics varied with age and gender. We also found increasing consultation frequency for some of the prodromal characteristics over time before diagnosis and heavier use of primary care services among people later diagnosed with psychosis than among those who did not develop psychosis. We have used these prodromal characteristics as candidate predictors for the development and validation of a prediction model for psychosis for primary care.

Approach

Study Design: Prospective cohort seeking help for mental health problems and a follow up of ≥5 years.Data source: Clinical Practice Research Datalink linked to Hospital Episode Statistics for diagnosis outcomes.Sample: 300,000 patients without previous coded diagnosis of psychosis but who consult for any other mental health problem.Primary outcome: ICD coded diagnosis of psychotic disorder. Period of risk is from first recorded consultation for a non-psychotic mental health problem until first recorded psychosis diagnosis.Predictors: Characteristics described above, age, gender and consultation frequency overall and per characteristic.Statistical Analysis: Odds ratios were calculated for each characteristic and entered into a logistic regression model. Area under ROC curve (AUROC) was calculated, together with sensitivity and specificity for exceeding the upper quartile of the risk distribution.

Findings

355 people were diagnosed with a psychotic disorder during follow up with a median follow-up of 7.9 (IQR 6.5, 8.9) years, resulting in an incidence of 1.5 per 10,000 person years (95% CI [1.4,1.7]) The most common diagnoses were; unspecified nonorganic psychosis (26.5%), delusional disorder (22.3%) and schizophrenia (16.6%). Of the cases 54% were female, mean age was 44 years, 88% were White and 33% were in the lowest deprivation quintile. Symptom prevalence ranged from 0.02% (bizarre behaviour) to 41.8% (depressive symptoms). Unadjusted odds ratios for the association between symptoms and outcome ranged from 36.5 (bizarre behaviour) to 1.01 (cannabis related problems). Prognostic scores integrating these predictors along with age and sex yielded a sensitivity 57.8%, specificity 75.2% with AUROC=0.7.

Consequences

These initial results suggest that the specified predictors are promising candidates for a risk score for psychosis and have higher predictive power than current psychosis risk prediction tools.