Composition of health care contact days and mortality prediction among trial participants.

454 Background: Health care contact days exceeding protocol-mandated contact days for trial participants may represent adverse clinical events and portend worse outcomes. Identifying patients at risk of poor outcomes could enable clinical teams to intervene early and support patients’ needs. We evaluated the association of contact days early in the trial course with mortality. Methods: We linked data from 6 SWOG trials [prostate (S9916, S0421), lung (S0003, S0124, S0819), and pancreatic (S0205) cancers] to Medicare claims. We calculated contact days (days with ambulatory, emergency department, inpatient, or facility-based care) and their composition using protocol calendars. Total contact days were delineated as protocol-mandated (planned) ambulatory vs unplanned (including unplanned ambulatory vs unplanned inpatient). Cox frailty landmark regression analysis was used in a random 60% training set to identify the optimal threshold (percentile) and landmark (months) for contact day measures as the predictor based on the chi-square statistic. Candidate predictors were tested in the remaining 40% sample. Results: We included 1,429 patients (median age, 71 years, 7.6% Black, 21.4% female, and 23.5% rural). In the first 3 months, among 15,301 contact days in the training set (6.4 contact days per person per month; 21.2% of total days), 4102 (26.8%) were protocol-mandated ambulatory, and 11,199 (73.2%) were unplanned [7328 (47.9%) unplanned ambulatory and 3871 (25.3%) unplanned inpatient]. The 2-month increment at the 75th percentile threshold was the optimal model in the training set. In the test set, unplanned contact days–especially unplanned inpatient contact days– was associated with subsequent survival (>2 unplanned inpatient contact days, adjusted HR=1.26 [1.02-1.56], p=.015) (Table). Conclusions: For trial participants, 1 of 5 follow-up days was a health care contact day. The majority of these were unplanned ambulatory contact days. Total unplanned contact days was marginally associated with subsequent survival. However, this association was driven by unplanned inpatient contact days, suggesting that unplanned ambulatory contact days may more represent uncoordinated care rather than care associated with future mortality. Meaningfully decreasing contact day burdens will require addressing uncoordinated unplanned ambulatory contact days (e.g., through navigation support). Unplanned inpatient contact days likely reflect more serious disease, and their low threshold (≥2 in the first 2 months) in predicting mortality highlights the adverse impact of acute care use on survival. Applying the optimal model to the testing set. Contact day predictor 2-month, 75 th percentile cut-point for contact days Adjusted HR (95% CI) p-value Total 16 1.12 (0.90-1.40) 0.16 Unplanned 11 1.20 (0.98-1.48) 0.04 Unplanned ambulatory 8 1.05 (0.87-1.28) 0.31 Unplanned inpatient 2 1.26 (1.02-1.56) 0.015