Optimal Matched Designs to Study Post-Acute Skilled Nursing Facility Utilization and Outcomes
View/ Open
Niknam_Dissertation_Submitted_Revised.pdf (4.809Mb)
Access Status
Full text of the requested work is not available in DASH at this time ("restricted access"). For more information on restricted deposits, see our FAQ.Author
Niknam, Bijan
Metadata
Show full item recordCitation
Niknam, Bijan. 2023. Optimal Matched Designs to Study Post-Acute Skilled Nursing Facility Utilization and Outcomes. Doctoral dissertation, Harvard University Graduate School of Arts and Sciences.Abstract
Medicare spends approximately $60 billion on post-acute care, with approximately half these expenditures being on Skilled Nursing Facility (SNF) care. As elsewhere in the healthcare system, racial inequities exist in SNF utilization, SNF characteristics, and post-acute outcomes. Using study designs that incorporate modern matching methods, we investigate sources of these inequities in an attempt to help policymakers pinpoint opportunities to most efficiently address them. More generally, there is extensive debate over the impact of SNF care on outcomes, which may be plausibly helpful or harmful depending on the patient in question. Because the selection process is uncontrolled, patients receiving SNF care are quite different from patients who do not, and because randomization to SNF care is unethical, researchers must rely on observational studies to study its effects. We contribute to this literature by introducing a new instrument for discharge to SNF care based on area SNF saturation, using matching to strengthen the instrument and facilitate sensitivity analysis and equivalence testing, and addressing residential selection on the instrument by concentrating on a cohort of traveling beneficiaries who were hospitalized outside their home state.Chapter 1: Examining Sources of Post-Acute Care Inequities with Layered Target Matching
The first chapter investigates inequities between Black and White patients in utilization of post-acute SNF care and subsequent outcomes. To parse the roles of demographics, patient characteristics, and hospitals in these inequities, we introduce a matching technique called layered target matching which blends the sequential adjustment framework of tapered matching with cardinality matching’s ability to form the largest balanced sample that is representative of a target population, which we specify in three layers as increasingly comprehensive sets of characteristics describing Black patients in the sample. We find that Black patients receive post-acute care in SNFs more frequently than White patients, yet despite this additional SNF care, they also experience significantly higher 30-day readmission rates. The layered matches show that differences in patient characteristics on admission explain some, but not all, of these inequities, with further adjustment for hospital characteristics having little effect. We also contrast the transparency of layered target matching with linear regression by characterizing the population implied by regression models, showing that unless done carefully, regression-based studies of inequities may be based on an implied population that is quite different from the vulnerable population under study, and hence may not accurately reflect the inequities these groups endure.
Chapter 2: Persistent Racial Inequities in Skilled Nursing Facility Care and Post-Acute Outcomes Despite Adjustment for Patient Characteristics, Geography, and Hospitals
As Black patients experienced higher readmission rates than White patients despite receiving additional SNF care, we next investigate differences in characteristics and quality of SNFs admitting Black and White patients, and how decision inputs such as the discharging hospital and geography influence patient allocation to SNFs and subsequent outcomes. Using a sample of Black and White patients who received post-acute SNF care, we conduct a series of increasingly comprehensive matches that first pair Black and White patients for 4,103 residential ZIP codes, 1,814 discharging hospitals, 2,313 combinations of discharging hospital and residential ZIP code, and finally 3,639 combinations of discharging hospital and admitting SNF, all while fixing patient characteristics to be representative of all Black patients. We found that Black patients tended to be admitted to larger SNFs and SNFs that were for-profit or chain members, with these differences declining but not disappearing after adjustment for discharging hospital and geography. While we observed substantial segregation of Black and White patients across SNFs, the gap in SNF racial composition fell by about half after matching exactly for the hospital and geography. 30-day readmissions remained persistently higher among Black patients regardless of the match, even when comparing Black and White patients who were discharged from the same hospital and admitted to the same SNF. Policymakers and hospital and SNF administrators must collaborate and potentially seek opportunities outside the admission episode to address inequitable readmission rates, for example through improving access to primary and preventive care to better equate risk on admission and thoroughness of follow-up outpatient care.
Chapter 3: Building a Stronger Instrument for Estimating the Impact of Skilled Nursing Facility Care on Post-Acute Outcomes
In Chapter 3, we estimate the effects of skilled nursing facility (SNF) care on post-acute outcomes using a new SNF concentration-based instrumental variables approach. We conceptualize a natural experiment using the hospital’s county-level SNF count per capita as an instrument for receiving post-acute SNF care. To counter bias due to residential selection, we concentrate on a sample of Medicare patients who were discharged from a hospital while traveling outside their home state. To further strengthen the instrument, improve balance, and facilitate sensitivity analysis, we use near-far cardinality matching to find the largest possible balanced sample of patients from higher-SNF lower-SNF areas who were balanced on 53 patient and hospital characteristics relative to each other as well as a target representative of all medical admissions in the national sample, while also being at least 2 standard deviations apart on mean county SNF count per capita. We then estimate the effects of SNF care on 30-day readmission and mortality using two-stage least-squares regression models and paired analyses. We found the instrument to be highly relevant to SNF discharge status, with rates significantly higher among traveling patients in counties with higher SNF concentrations. This may be evidence of supplier-induced demand for post-acute SNF care that is plausibly independent of clinical need. Examining outcomes, although ordinary least-squares regression models showed discharge to a SNF was associated with significantly lower readmission and mortality, two-stage least-squares models on the matched sample found no significant risk or benefit for readmission and somewhat higher mortality within 30 days of discharge, though this effect did not reach statistical significance. Finally, by contrasting results across four approaches that incorporate or omit considerations for clinical and residential selection processes in their designs, we show that approaches that do not do so may yield misleading conclusions.
Terms of Use
This article is made available under the terms and conditions applicable to Other Posted Material, as set forth at http://nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of-use#LAACitable link to this page
https://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37375695
Collections
- FAS Theses and Dissertations [6136]
Contact administrator regarding this item (to report mistakes or request changes)