In this article, we describe twopm, a command for fitting two-part models for mixed discrete-continuous outcomes. In the two-part model, a binary choice model is fit for the probability of observing a positive-versus-zero outcome. Then, conditional on a positive outcome, an appropriate regression model is fit for the positive outcome. The twopm command allows the user to leverage the capabilities of predict and margins to calculate predictions and marginal effects and their standard errors from the combined first- and second-part models.

Health care expenditures and use are challenging to model because these dependent variables typically have distributions that are skewed with a large mass at zero. In this article, we describe estimation and interpretation of the effects of a natural experiment using two classes of nonlinear statistical models: one for health care expenditures and the other for counts of health care use. We extend prior analyses to test the effect of the ACA’s young adult expansion on three different outcomes: total health care expenditures, office-based visits, and emergency department visits. Modeling the outcomes with a two-part or hurdle model, instead of a single-equation model, reveals that the ACA policy increased the number of office-based visits but decreased emergency department visits and overall spending. Expected final online publication date for the Annual Review of Public Health Volume 39 is April 1, 2018. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.

Objective: To compare per diem total direct, ancillary (laboratory and radiology) and pharmacy costs of palliative care (PC) compared to usual care (UC) patients during a terminal hospitalization; to examine the association between PC and ICU admission. Design: Retrospective, observational cost analysis using a VA (payer) perspective. Setting: Two urban VA medical centers. Measurements: Demographic and health characteristics of 314 veterans admitted during two years were obtained from VA administrative data. Hospital costs came from the VA cost accounting system. Analysis: Generalized linear models (GLM) were estimated for total direct, ancillary and pharmacy costs. Predictors included patient age, principal diagnosis, comorbidity, whether patient stay was medical or surgical, site and whether the patient was seen by the palliative care consultation team. A probit regression was used to analyze probability of ICU admission. Propensity score matching was used to improve balance in observed covariates. Results: PC patients were 42 percentage points (95% CI, –556% to –31%) less likely to be admitted to ICU. Total direct costs per day were $239 (95% CI, –387 to –122) lower and ancillary costs were $98 (95% CI, –133 to –57) lower than costs for UC patients. There was no difference in pharmacy costs. The results were similar using propensity score matching. Conclusion: PC was asssociated with significantly lower likelihood of ICU use and lower inpatient costs compared to UC. Our findings coupled with those indicating better patient and family outcomes with PC suggest both a cost and quality incentive for hospitals to develop PC programs.

Despite its demonstrated potential to both improve quality of care and lower costs, the Medicare hospice benefit has been seen as producing savings only for patients enrolled 53–105 days before dea...