Population mental health improves with increasing access to treatment: evidence from a dynamic modelling analysis

Abstract

Background: Multiple studies indicate that the prevalence of mental disorders in high-income countries has remained stable or increased despite substantial increases in the provision of care, leading some authors to question the effectiveness of increasing access to current treatments as a means of improving population mental health.

Methods: We developed a system dynamics model of mental disorder incidence and treatment-dependent recovery to assess two potential explanations for the apparent failure of increasing treatment provision to reduce mental disorder prevalence: 1) an increase in the individual-level risk of disorder onset; and 2) declining effectiveness of care resulting from insufficient services capacity growth. Bayesian Markov Chain Monte Carlo (MCMC) methods were used to fit the model to data on the prevalence of high to very high psychological distress in Australia for the period 2008-2019.

Results: Estimates of yearly rates of increase in the per capita incidence of high to very high psychological distress and the proportion of patients recovering when treated indicate that the individual-level risk of developing high to very high levels of distress increased between 2008 and 2019 (posterior probability > 0.999) but provide no evidence for declining treatment effectiveness. Simulation analyses suggest that the prevalence of high to very high psychological distress would have decreased from 14.4% in 2008 to 13.6% in 2019 if per capita incidence had not increased over this period (prevalence difference 0.0079, 95% credible interval 0.0015-0.0176).

Conclusions: Our analyses indicate that a modest but significant effect of increasing access to mental health care in Australia between 2008 and 2019 was obscured by a concurrent increase in the incidence of high to very high psychological distress.

Keywords: Australia; Bayesian analysis; Mental health services; Psychological distress; System dynamics.

Fig. 1

System dynamics model of psychological distress incidence and treatment-dependent recovery used for the analyses. Notation is defined in the Methods section and Table 2. Stocks (or state variables) are shown as boxes, flows as pipes with taps, causal connections (or mathematical dependencies) as arrows, and sources and sinks as clouds [15]. Symbols with dashed outlines are copies (or ‘ghosts’) of the corresponding symbols with solid outlines

Fig. 2

Left panels. Marginal posterior distributions estimated for the fractional rate of increase in the per capita incidence of high to very high psychological distress (δ_i) and the fractional rate of increase in treatment effectiveness (δ_r). Median estimates and 95% credible intervals are shown in the top right corner of each panel. Prior distributions are plotted as smooth curves. The close similarity of the posterior and prior distributions for δ_r indicates that the available data provide no evidence for declining (or increasing) treatment effectiveness (the prior is symmetrical about 0). Right panels. Modelled trajectories for the per capita incidence of high to very high psychological distress (i) and the proportion of patients recovering with treatment (r) over the period 2008 to 2020. Pointwise 50 and 95% credible intervals (calculated from the output of 10³ simulations, each using a randomly selected parameter vector θ sampled in the Markov chain Monte Carlo analysis) are indicated with dark grey shading and light grey shading, respectively

Fig. 3

A Estimates of the prevalence of high to very high psychological distress among Australian adults (18 years and above) over the period 2008 to 2020 derived from the Household, Income and Labour Dynamics in Australia (HILDA) Survey (red open circles with 95% confidence intervals; see ref. [19]) and the system dynamics model (dark grey line, obtained assuming median parameter estimates). Pointwise 50 and 95% credible intervals (calculated from the output of 10³ simulations, each using a randomly selected parameter vector θ sampled in the Markov chain Monte Carlo analysis) are indicated with dark grey shading and light grey shading, respectively. B Prevalence of high to very high psychological distress simulated under a counterfactual scenario in which per capita incidence remains constant over time (δ_i is set to 0; red line). The model-based estimates from panel A (where δ_i is estimated from the HILDA Survey data) are also plotted for comparison (dark grey line). Pointwise 50 and 95% credible intervals are indicated with dark shading and light shading, respectively

Fig. 4

Estimates of the prevalence of high to very high psychological distress among Australian adults (18 years and above) derived from the Household, Income and Labour Dynamics in Australia (HILDA) Survey (red open circles with 95% confidence intervals) and from constrained models in which the fractional rates of increase in the per capita incidence of high to very high psychological distress and treatment effectiveness (δ_i and δ_r, respectively) are assumed to equal 0 (dark grey lines). Pointwise 50 and 95% credible intervals are indicated with dark grey shading and light grey shading, respectively