Use of Clinical Notes to Assess Neuropsychiatric Events After Montelukast Initiation

TL;DR

Using clinical notes along with structured data improves detection of neuropsychiatric events linked to montelukast use.

Contribution

Demonstrates that combining claims and unstructured clinical notes enhances outcome identification in drug safety studies.

Findings

Adding clinical notes identified 20% more patients with neuropsychiatric events compared to structured data alone.

Hazard ratios for neuropsychiatric events decreased as more data sources were added.

Drug safety studies should integrate clinical notes for more complete evidence.

Abstract

What is the value of including information extracted from unstructured data in a drug safety study on the association of montelukast initiation and risk of neuropsychiatric events?

In this cohort study of 109 076 patients, 20% more patients with outcomes associated with neuropsychiatric events were identified when using clinical notes compared with structured data alone.

These findings suggest that drug safety studies able to use all available health care data sources, including claims and with both structured and unstructured electronic health care data, may provide more complete and granular ascertainment of study variables and lead to more robust and precise findings.

This cohort study examines the value associated with using structured and unstructured clinical notes linked with claims data while investigating the risk of neuropsychiatric disorders among montelukast users.

Prior drug safety studies investigating the risk of neuropsychiatric events (NPEs) after montelukast initiation have been limited by methodological issues, including incomplete confounder control and unmeasured outcomes associated with completeness of structured data, as well as events recorded only in unstructured clinical notes.

To examine the value associated with using linked claims and electronic health records (EHRs) (structured and unstructured data) while investigating the risk of any NPE among patients with asthma initiating montelukast compared with inhaled corticosteroids (ICSs).

This retrospective cohort study used Oracle EHR Real-World Data linked to a national US claims dataset (July 1, 2015, to June 30, 2022) to identify patients aged 6 to 80 years with asthma newly initiating montelukast or ICSs. The data were analyzed between December 13, 2022, and August 2, 2024.

The primary outcome of any NPE and covariates was assessed using the incremental addition of each data source: analysis 1, claims-only data; analysis 2, claims plus structured EHR data; and analysis 3, claims plus structured and unstructured EHR data. Cox proportional hazard regression models using propensity score–matched cohorts across data sources were used to examine the risk of any NPE by treatment initiation group.

Among 109 076 patients (mean [SD] age at treatment initiation, 28.8 [20.5] years, 59.4% female), 39 665 (36.4%) initiated montelukast and 69 411 (63.6%) ICSs. Incidence rates per 100 person-years of the first postindex NPE increased with additional data sources for both montelukast and ICS users (analysis 1, 17.11 [95% CI, 16.86-17.36] vs 15.57 [95% CI, 15.34-15.80], respectively; analysis 2, 19.10 [95% CI, 18.83-19.38] vs 18.23 [95% CI, 17.97-18.50], respectively; analysis 3, 27.78 [95% CI, 27.43-28.13] vs 27.40 [95% CI, 27.06-27.75], respectively). Hazard ratios were attenuated across contributing data sources for analysis 1 (1.08 [95% CI, 1.05-1.11]), analysis 2 (1.04 [95% CI, 1.01-1.06]), and analysis 3 (1.01 [95% CI, 1.00-1.03]).

This cohort study found that clinical information from linked claims and structured and unstructured EHR data was associated with enriched measurement of patient and disease characteristics and enhanced completeness of evidence vs claims data alone. The findings, however, did not differ substantially across the incrementally contributing data sources or from prior studies. Drug safety and effectiveness studies should integrate information using clinical notes from EHRs with consideration of potential limitations, challenges, and necessary validation processes.

Figures