Table 3. Association of Medicare Advantage (MA) vs Traditional Medicare (TM) With Ambulatory Care Access and Quality for Beneficiaries With Disability Entitlement, 2015-2018.
| Variable | Unadjusted results | Adjusted marginal difference of MA vs TM, (95% CI) | |||
|---|---|---|---|---|---|
| MA | TM | Absolute difference (95% CI) | Regression resultsa | Propensity-weighted regression resultsb | |
| Accessc | |||||
| Usual source of care, % | 90.2 | 84.9 | 5.3 (3.2 to 7.4) | 3.7 (1.5 to 5.9) | 2.9 (0.2 to 5.7) |
| Usual source of care is PCC, % | 77.4 | 70.1 | 7.2 (3.5 to 11.0) | 5.0 (1.1 to 9.0) | 3.0 (– 0.8 to 6.8) |
| Specialist visit, % | 53.2 | 44.8 | 8.3 (4.3 to 12.3) | 4.9 (0.7 to 9.0) | 5.5 (0.6 to 10.5) |
| Quality | |||||
| Annual cholesterol screen, %d | 91.1 | 86.4 | 4.7 (1.7 to 7.8) | 3.5 (0.8 to 6.2) | 3.8 (0.9 to 6.7) |
| Annual flu shot, %e | 61.4 | 51.5 | 9.9 (6.0 to 13.8) | 10.1 (5.3 to 14.8) | 10.4 (5.3 to 15.5) |
| Colon cancer screening, %f | 68.4 | 54.6 | 13.8 (9.3 to 18.3) | 11.5 (6.4 to 16.5) | 10.3 (4.8 to 15.8) |
Abbreviation: PCC, primary care clinician.
We estimated multivariable logistic regression models for each outcome that also adjusted for the characteristics listed in Table 2 (with race and ethnicity collapsed into minority vs other). We added fixed effects for the states that beneficiaries resided in to control for state policy differences and state differences in supply of medical services, clinician practice intensity, and coding intensity. We included year fixed effects to control for secular trend and adjusted our P values for the complex survey design of the Medicare Current Beneficiary Survey and intra-person correlation over time. We used Stata's Margins command to report our results as the marginal difference of MA vs TM for the dependent variables by modeling the response in the dependent variables to the exposure variable at the population means.
We estimated the same multivariable logistic regression models as in a, but this time reweighting the sample using the propensity score weights described previously to change the distribution of observed confounders in both the treated (MA) and untreated (TM) beneficiaries so that they are the same as the distribution in the entire sample. These estimates should be interpreted as what we would expect to see if every Medicare beneficiary in our nationally representative sample enrolled in MA vs what we would expect to see if nobody enrolled in MA (ie, the average treatment effects).
Unweighted sample n = 6525. Met baseline study inclusion and responded to Medicare Current Beneficiary Survey questions for outcome variables.
Unweighted sample n = 2715. Met baseline study inclusion and exclusion criteria and self-reported having diabetes, ischemic heart disease, or heart failure and responded to Medicare Current Beneficiary Survey questions for outcome variable.
Unweighted sample n = 6462. Met baseline study inclusion and exclusion criteria and responded to Medicare Current Beneficiary Survey question for outcome variable.
Fecal occult blood test at home or physician’s office or colonoscopy or sigmoidoscopy within past 5 years, excluding patients who self-reported having colon cancer or were younger than 45 years. Unweighted sample n = 3233 for patients who met above criteria as well as baseline study inclusion and exclusion criteria and responded to Medicare Current Beneficiary Survey questions for outcome variable.