Mayer SE, Tan HJ, Peacock Hinton S, Hester LL, Sturmer T, Faurot KR, Jonsson Funk M, Sanoff HK, Lund JL. Medicare claims-based measures of poor physical function and associations with treatment and mortality in older colon cancer patients. Poster presented at the 33rd ICPE International Conference on Pharmacoepidemiology & Therapeutic Risk Management; August 26, 2017. Montreal, Canada. [abstract] Pharmacoepidemiol Drug Saf. 2017 Aug 22; 26(Suppl 2):174. doi: 10.1002/pds.4275


BACKGROUND: Non-experimental studies using Medi-care data to evaluate drug effects in older adults are of-ten subject to unmeasured confounding by functionalstatus. Three research groups have developed Medi-care claims-based proxies for poor physical function(Faurot (F), Davidoff (D), and Chrischilles (C)); how-ever, no studies have applied and compared these mea-sures within a single cohort.

OBJECTIVES: This study evaluates agreement betweenthe three measures and associations with treatmentand mortality in older colon cancer patients, whereconfounding by functional status is likely.

METHODS: Medicare beneficiaries diagnosed withstage II/III colon cancer undergoing surgical resectionwere identified in the Surveillance, Epidemiology, andEnd Results-Medicare data (2004–2011). All patientshad continuous Medicare Parts A/B coverage for 12+months before diagnosis to define the claims-basedmeasures. Each model included 16 indicators of poorfunction. Poor function was operationalized and com-pared by 1) summing the number of indicators fromF, D, and C separately and 2) estimating the predictedprobability of poor function from F and D only.Agreement was evaluated using kappa and Pearsoncorrelation coefficients. Associations between eachclaims-based measure and 1) adjuvant chemotherapyreceipt and 2) mortality were estimated using log-bino-mial and Cox proportional hazards regression models,controlling for age, sex, stage, and comorbidity score.

RESULTS: Of the 29,687 patients, 67% were age 75+and 45% had stage III cancer. The proportion of pa-tients with 3+ indicators ranged from 9% (C) to 24%(F); median predicted probability of poor function dif-fered by measure (F: 0.05 (IQR: 0.03, 0.12); D: 0.02(IQR: 0.01, 0.04)). Concordance across the three indi-cator counts was low (weighted kappa: 0.35–0.39),while predicted probabilities of poor function weremoderately correlated (Pearson’sr= 0.63). Higherpredicted probability of poor function (>10% vs<5%) was associated with lower risk of adjuvant che-motherapy receipt (F: adjusted risk ratio (aRR) = 0.61(0.57, 0.65); D: aRR = 0.67 (0.63, 0.72)) and highermortality (F: aHR = 1.99 (1.89, 2.10); D: aHR =1.62 (1.53, 1.70)).

CONCLUSIONS: While the three measures were notstrongly correlated, each was associated with treat-ment and mortality. Future efforts to combine thesemodels may improve control for confounding by func-tional status in non-experimental studies of olderadults using administrative data.

Share on: