Abstract 3303: Development and initial validation of a prognostic score based on structured data from EHRs

Introduction Real world data (RWD) is increasingly used to supplement clinical trial data. Prognostic scores (PS) such as ECOG performance status (ECOG) are important for matching patient populations and assessing treatment impact, yet such PS are often missing from patient records or are embedded i...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Cancer research (Chicago, Ill.) Ill.), 2019-07, Vol.79 (13_Supplement), p.3303-3303
Hauptverfasser: Backenroth, Daniel, Haimson, Josh D., Meropol, Neal J., Baxi, Shrujal S.
Format: Artikel
Sprache:eng
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Introduction Real world data (RWD) is increasingly used to supplement clinical trial data. Prognostic scores (PS) such as ECOG performance status (ECOG) are important for matching patient populations and assessing treatment impact, yet such PS are often missing from patient records or are embedded in physician narratives. An accurate PS that does not rely on physician assessment or data entry would have broad utility. We therefore undertook development of a PS based solely on structured data from RWD across numerous cancer diagnoses (the structured PS). We then sought initial validation of the structured PS in patients treated with checkpoint inhibitors (CIT). Methods Using the Flatiron Health electronic health record-derived database, we applied LASSO regression to develop the structured PS to predict for overall survival (OS) after starting any first-line (1L) cancer therapy. Data included laboratory values and vital signs from between 30 days before and 15 days after start of treatment as well as age. The structured PS was developed from ~39K patients with multiple myeloma, chronic lymphocytic leukemia or one of 8 solid tumor malignancies. We evaluated the structured PS using concordance, i.e., how often the score correctly predicts the relative OS of two randomly selected patients, on ~24K patients, comparing to a baseline based on age and ECOG, when available. We then measured the association of OS with the structured PS at start of 1L CIT in patients with five tumor types where 1L CIT is common. Results The concordance of the structured PS was 66%-73%, on average 6% higher than that of age and ECOG. Table 1 shows by cancer type # of patients and median OS from 1L CIT for quartiles of the structured PS (NR=not reached). Conclusion A structured PS can be an important pan-tumor variable in investigations of OS using RWD, including for patients treated with first-line CIT. OS from 1L CIT by quartiles of structured PSDiseasenQ1 median (95% CI) in monthsQ2 median (95% CI) in monthsQ3 median (95% CI) in monthsQ4 median (95% CI in months)Head and neck3704 (3-5)9 (7-12)14 (12-Inf)NR (NR-Inf)Melanoma16355 (4-6)20 (15-26)33 (22-Inf)NR (NR-Inf)Non-small cell lung cancer41714 (3-4)9 (8-10)14 (13-16)24 (21-29)Renal cell carcinoma2775 (3-7)23 (12-Inf)20 (13-Inf)NR (28-Inf)Urothelial cancer6122 (2-3)6 (4-7)9 (7-13)NR (15-Inf) Note: This abstract was not presented at the meeting. Citation Format: Daniel Backenroth, Josh D. Haimson, Neal J. Meropol, Shrujal S. Baxi. D
ISSN:0008-5472
1538-7445
DOI:10.1158/1538-7445.AM2019-3303