Pseudorandomization using an instrumental variable: a strong tool to break through selection bias

Comparing Suicide Rates for Cognitive Processing Therapy Versus Prolonged Exposure Therapy for Posttraumatic Stress Disorder

OBJECTIVE

This study aimed to compare suicide mortality rates for patients receiving two evidence-based psychotherapy (EBP) protocols for posttraumatic stress disorder (PTSD): cognitive processing therapy (CPT) and prolonged exposure (PE).

METHODS

Suicide mortality was measured among U.S. Department of Veterans Affairs patients with PTSD who received EBP from 2009 through 2019. Regional variation in delivering CPT versus PE was leveraged as an instrumental variable (IV) to compare suicide mortality by using standard adjustment and IV-based analyses.

RESULTS

In total, 62,686 patients received EBP for PTSD; 82.4% were male, and the mean±SD age was 46.9±14.4. Patients were followed for a median of 6 years, and there were 136 deaths by suicide (38.3 and 32.4 per 100,000 person-years among the CPT and PE groups, respectively). The regional rate of CPT versus PE delivery was a strong IV that had greater explanatory power for the type of EBP received than all patient factors combined. The standard adjustment model for CPT produced a hazard ratio of 1.25, whereas the reduced-form IV produced a hazard ratio of 1.22. The probit IV, in which relevant covariates were updated annually, produced an odds ratio of 0.99. The time-to-event IV produced a hazard ratio of 1.20. The differences were not significant.

CONCLUSIONS

No statistically significant difference was found between CPT and PE in the outcome of death by suicide. More effective interventions that result in higher remission rates would likely need to be developed to achieve a relative decrease in suicide risk through PTSD treatment.

Liver Resection Improves Survival in Colorectal Cancer Patients: Causal-effects From Population-level Instrumental Variable Analysis

OBJECTIVE

The aim of this study was to estimate population-level causal effects of liver resection on survival of patients with colorectal cancer liver metastases (CRC-LM).

BACKGROUND

A randomized trial to prove that liver resection improves survival in patients with CRC-LM is neither feasible nor ethical. Here, we test this assertion using instrumental variable (IV) analysis that allows for causal-inference by controlling for observed and unobserved confounding effects.

METHODS

We abstracted data on patients with synchronous CRC-LM using the California Cancer Registry from 2000 to 2012 and linked the records to the Office of Statewide Health Planning and Development Inpatient Database. We used 2 instruments: resection rates in a patient's neighborhood (within 50-mile radius)-NALR rate; and Medical Service Study Area resection rates-MALR rate. IV analysis was performed using the 2SLS method.

RESULTS

A total of 24,828 patients were diagnosed with stage-IV colorectal cancer of which 16,382 (70%) had synchronous CRC-LM. Liver resection was performed in 1635 (9.8%) patients. NALR rates ranged from 8% (lowest-quintile) to 11% (highest-quintile), whereas MALR rates ranged from 3% (lowest quintile) to 19% (highest quintile). There was a strong association between instruments and probability of liver resection (F-statistic at median cut-off: NALR 24.8; MALR 266.8; P < 0.001). IV analysis using both instruments revealed a 23.6 month gain in survival (robust SE 4.4, P < 0.001) with liver resection for patients whose treatment choices were influenced by the rates of resection in their geographic area (marginal patients), after accounting for measured and unmeasured confounders.

CONCLUSION

Less than 10% of patients with CRC-LM had liver resection. Significant geographic variation in resection rates is attributable to community biases. Liver resection leads to extensive survival benefit, accounting for measured and unmeasured confounders.