Developing prognostic models for health care utilization in patients with work-related mental health problems

Abstract

Background: The long-term prognosis for employees with work-related mental health problems is unclear. We aim to describe long-term trends in health care utilization (HCU) and develop multivariable prognostic models for long-term mental health care utilization. Methods: From the Danish Occupational Medicine Cohort we included mental health patients (N = 17,822) assessed from 2000 to 2013 at Departments of Occupational Medicine. Outcomes were general health (general practitioner, somatic hospital) and mental health (psychiatrist/psychologist, psychiatric hospital) HCU obtained from registries five years before/after assessment. The 10-year period was divided into phases relative to assessment: 5 − 3 years before, 2 years before/after, and 3–5 years after. We developed gender-stratified Lasso-penalized multivariable prognostic models for HCU 3–5 years after assessment assessing both calibration and discrimination. Results: Prevalent HCU for general practitioner, psychiatrist/psychologist and psychiatric hospital services was relatively stable 5 − 3 years prior to assessment, then rising during the 2 years before/after. At 3–5 years after assessment prevalent general practitioner HCU declined to previous levels, while prevalent HCU for psychologist/psychiatrist and psychiatric hospital services remained elevated compared to previous levels during years 5 − 3. Prognostic models for long-term psychologist/psychiatrist and psychiatric hospital HCU indicated acceptable calibration and modest discrimination. Conclusions: Prevalent HCU rose two years before/after assessment and remained elevated for psychiatrist/psychologist and psychiatric hospital HCU 3–5 years after. Gender-stratified prognostic models were developed for long-term mental health HCU, but discrimination and calibration should be further improved before out-of-sample application for personal prognosis. Trial registration: The study was registered at clinicaltrials.gov (Identifier: NCT04459793) prior to analyses.