Creating a Proxy for Baseline Eastern Cooperative Oncology Group Performance Status in Electronic Health Records for...

Area of Science:

• Oncology
• Health Informatics
• Biostatistics

Background:

• Eastern Cooperative Oncology Group performance status (ECOG PS) is crucial for cancer research but often missing in electronic health records (EHRs).
• Incomplete ECOG PS data can bias comparative effectiveness research by introducing survival differences between patients with known and imputed data.
• Existing imputation methods may not adequately preserve clinical outcomes, complicating research.

Purpose of the Study:

• To develop and validate a novel imputation method for missing ECOG PS in advanced non-small cell lung cancer (aNSCLC) patients.
• To ensure that patients with imputed ECOG PS are indistinguishable from those with known ECOG PS in terms of survival outcomes.
• To reduce potential biases in comparative effectiveness research stemming from incomplete ECOG PS data.

Main Methods:

• Utilized a deidentified EHR database of aNSCLC patients undergoing first-line treatment.
• Employed a novel imputation strategy involving sample-splitting, predictive modeling, and an optimal threshold determined by clinical outcomes.
• Validated the approach using classification metrics and by comparing survival metrics between observed and imputed ECOG PS.

Main Results:

• 21% of 62,101 patients had missing ECOG PS at baseline.
• The novel imputation method demonstrated comparable or superior accuracy (73.3%), sensitivity (42.4%), and specificity (81%) versus other techniques.
• Achieved minimal survival metric differences: 0.07 hazard ratio, -0.36 months median survival for good ECOG PS (<2), and -0.39 months for poor ECOG PS (≥2).

Conclusions:

• The developed imputation method effectively aligns clinical outcomes, enhancing the utility of EHR data for aNSCLC research.
• This approach improves the reliability of comparative effectiveness studies by minimizing bias from missing ECOG PS data.
• The method offers a robust solution for leveraging real-world data in oncology research.