Combining Machine Learning and Comparative Effectiveness Methodology to Study Primary Care Pharmacotherapy Pathways...

Area of Science:

• Pharmacotherapy and Comparative Effectiveness Research
• Machine Learning in Healthcare Analytics
• Mental Health Treatment Optimization

Background:

• Current antidepressant prescribing patterns often follow a "start low, go slow" approach.
• The effectiveness of different antidepressant titration strategies on patient engagement remains understudied.
• Major depressive disorder treatment requires sustained patient engagement for optimal outcomes.

Purpose of the Study:

• To introduce an innovative methodology combining machine learning and comparative effectiveness research.
• To investigate the impact of antidepressant prescribing patterns on patient engagement in care.
• To evaluate the effectiveness of different antidepressant dose escalation strategies.

Main Methods:

• Utilized United States Veterans Health Administration data from 2006-2020.
• Applied process mining and machine learning to generate pharmacotherapy pathways for antidepressants.
• Employed 2-stage least squares with instrumental variables (provider practice patterns) to assess dose escalation strategies, controlling for patient and provider characteristics.

Main Results:

• A statistically significant positive effect (0.68) of rapid dose escalation ("ramping up fast") on engagement in care was observed.
• Slow dose escalation ("ramping up slow") showed an insignificant negative impact (-0.82) on engagement.
• Higher probability of dropout negatively impacted engagement (-0.39); results were validated using medication possession ratios.

Conclusions:

• Findings challenge the traditional "start low, go slow" antidepressant dosing strategy.
• Faster antidepressant dose escalation demonstrates a significantly positive effect on patient engagement.
• This approach may improve treatment adherence and outcomes in patients with major depressive disorder.