Currently Viewing:
The American Journal of Managed Care April 2016
Single- Versus Multiple-Tablet HIV Regimens: Adherence and Hospitalization Risk
S. Scott Sutton, PharmD; James W. Hardin, PhD; Thomas J. Bramley, RPh, PhD; Anna O. D'Souza, BPharm, PhD; and Charles L. Bennett, MD, PhD, MPP
Medicaid Managed Care Reduces Readmissions for Youths With Type 1 Diabetes
Kathleen Healy-Collier, CSSBB, DHA; Walter J. Jones, PhD; James E. Shmerling, DHA, FACHE; Kenneth R. Robertson, MD, MBA; and Robert J. Ferry, Jr, MD, FAAP
Currently Reading
Assessing the Impact of an Integrated Care System on the Healthcare Expenditures of Children With Special Healthcare Needs
Mircea I. Marcu, PhD; Caprice A. Knapp, PhD; David Brown, PhD; Vanessa L. Madden, BSc; and Hua Wang, MS
Lost in Translation: Healthcare Utilization by Low-Income Workers Receiving Employer-Sponsored Health Insurance
Bruce W. Sherman, MD; Wendy D. Lynch, PhD; and Carol Addy, MD, MMSc
Patient Safety Intervention to Reduce Unnecessary Red Blood Cell Utilization
Scott Hasler, MD; Amanda Kleeman MS; Richard Abrams, MD; Jisu Kim, MD; Manya Gupta, MD; Mary Katherine Krause, MS; and Tricia J. Johnson, PhD
Impact of Clinical Pharmacy Services on Outcomes and Costs for Indigent Patients With Diabetes
Marissa Escobar Quinones, PharmD, CDE; Margaret Youngmi Pio, PharmD, BCPS, CDE; Diem Hong Chow, PharmD, CDE; Elizabeth Moss, PharmD, CDE, BCACP; Jeffrey Lynn Hulstein, PharmD, CDE; Steven Micheal Boatright, PharmD, CDE; and Annie Mathew, PharmD, CDE
Adding Glucose-Lowering Agents Delays Insulin Initiation and Prolongs Hyperglycemia
Courtney Hugie, PharmD, BCPS; Nancee V. Waterbury, PharmD, BCACP; Bruce Alexander, PharmD; Robert F. Shaw, PharmD, MPH, BCPS, BCNSP; and Jason A. Egge, PharmD, MS, BCPS
Costs for a Health Coaching Intervention for Chronic Care Management
Todd H. Wagner, PhD; Rachel Willard-Grace, MPH; Ellen Chen, MD; Thomas Bodenheimer, MD, MPH; and David H. Thom, MD, PhD, MPH
Four Steps for Improving the Consumer Healthcare Experience Across the Continuum of Care
Keith Roberts, MBA
Patient Perceptions of Clinician Self-Management Support for Chronic Conditions
Peter Cunningham, PhD

Assessing the Impact of an Integrated Care System on the Healthcare Expenditures of Children With Special Healthcare Needs

Mircea I. Marcu, PhD; Caprice A. Knapp, PhD; David Brown, PhD; Vanessa L. Madden, BSc; and Hua Wang, MS
This study analyzes the effect of a managed care program on Medicaid expenditures for children with special healthcare needs using a quasi-experimental design.
ln(yit) = Zit' β+εit= Xit' γ+α×Countyi+δ×Timet+θ×Countyt×Timet+εit.               (2)

The explanatory variables contain confounding factors. The impact of the ICS implementation is captured in the interaction term “county × time,” which equals 1 for children in a treatment county after the implementation of the ICS, and 0 otherwise.

Our interest was in the impact of policy and explanatory variables on dollar costs, rather than on the log scale. This created the necessity for retransforming the results onto the dollar scale:


We estimated 3 log OLS models that differed in the way the log results were retransformed:

Normal Distribution. Assuming that the disturbances from the log OLS regression are normally distributed with mean 0 and variance σ2, we have:


Duan’s Smear. If the errors εit are homoscedastic, the last term in equation (3) is consistently estimated by the smearing factor proposed by Duan even if the errors are not normally distributed21:

, where          


Ai-Norton. Ai and Norton proposed an alternative, semi-parametric method for retransforming the results on the scale of interest when the errors are heteroscedastic.16 Their method is based on modeling the expectation of the exponentiated residuals from the log regression using a polynomial approximation and the predicted values from this regression:



We used a polynomial approximation Wit in which we included the same variables as in Zit, as well as higher-order terms of age, together with an interaction between these terms and gender.

Generalized Linear Models

GLMs specify a function between the index Zit' β and the expected value of the cost variable of interest, and a distribution that reflects the mean-variance relationship in the data.

GLM with gamma distribution. We used the generalized gamma distribution, which subsumes the normal, exponential, and Weibull distributions as particular cases. We used the log link function:


The GLM approach performs better than the log transform when the data are severely skewed and log costs are not symmetrically distributed.14

Extended generalized linear model (EGLM). Alternative link functions can be used instead of the log link. Box-Cox estimated parameters are close to 0, but statistically significant for all our cost variables of interest, indicating that, although the log link is appropriate, more efficient power transformations may exist. An alternative EGLM framework to simultaneously estimate the parameter λ of a Box-Cox transformation with those of a GLM model was used14:


where the variance is assumed to be:


with               .                      (10)

Impact of ICS

For each of the 2-part models and the cost categories of interest, we estimated the impact of the ICS on each cost category by the average treatment effect on the treated (ATET). This reflects the overall impact of the ICS program in a given reform county. Using the ATET allowed us to compare the treatment effect of the ICS program across the various econometric models considered.

Table 1 presents the characteristics of children who resided in Broward, Duval, and the corresponding control counties in our sample. Table 1 reveals that various characteristics of the children in the treatment group and corresponding control group are not statistically different, pre-treatment, for both counties. However, the treatment counties have a higher proportion of black non-Hispanics, a lower proportion of Hispanics, a lower average age, and more children categorized as disabled. These important characteristics are included in the regression analyses as covariates to control for these statistically significant differences.

The average monthly costs in each county and period are presented in Table 2. The average monthly costs in each category were higher in Broward than in its control county both before and after the implementation of the ICS, whereas they were lower in Duval than in its corresponding control county. The average monthly costs in each category were lower after the beginning of the implementation of the ICS in each county, with the exception of pharmacy costs in the Duval control. As in most healthcare cost studies, all the cost variables were skewed to the right.

eAppendix Table 1 (eAppendices available at shows that the EGLM method proposed systematically had the lowest RMSE among the models we estimated for each of the cost categories.14 The log OLS model with the retransformation proposed by Ai and Norton performed best among the outpatient models we estimated for Broward and its control county.16 The log OLS with normal retransformation had the highest in-sample accuracy among the total costs models estimated for Duval and its control county.

Table 3 presents the estimation results of the most accurate (lowest in-sample RMSE) 2-part models for each cost category of interest for Broward and its control, and Table 4 presents the models with the lowest in-sample RMSE for Duval and its control county. Focusing on the ICS variable in the second part of the 2-part models for Duval and its control, the ICS coefficients are negative and statistically significant for the total, inpatient, outpatient, and pharmacy costs. This suggests that the ICS program reduced costs in these categories in Duval County. Alternatively, using Table 3, outpatient, inpatient, pharmacy, and total costs were also lower in Broward County after the implementation of the ICS, but the estimated cost decreases were not statistically significant.

We estimated the ATET, which represents the sample average of the estimated dollar effect of the ICS on the healthcare costs of children in each of the ICS counties (eAppendix Table 2). The implementation of the ICS in Duval County appears to have led to economically and statistically significant outpatient cost savings ($84 and $1008 per CMSN enrollee per month and per year, respectively) and total cost savings ($195 and $2340 per CMSN enrollee per month and per year, respectively).

The lowest RMSE model indicates an average monthly total cost reduction of about $188 per CMSN enrollee per month ($2256 per year) in Broward due to the implementation of the ICS. All ATET estimates for Broward and its control are negative except the ones for average monthly ED costs. However, none of the dollar estimates for Broward is statistically significant at the 5% level.

The literature on the effect of enrollment in managed care models on healthcare expenditures of CSHCN is limited. Our study aimed to address this gap by using a recent policy change to describe the effects of an ICS on the healthcare expenditures of CSHCN in Florida. Our study is unique, in that Florida did not carve out specific services for CSHCN, but chose to carve out the entire system of healthcare. Additionally, the ICS ushered in a level of managed care that was not previously experienced. 

Copyright AJMC 2006-2019 Clinical Care Targeted Communications Group, LLC. All Rights Reserved.
Welcome the the new and improved, the premier managed market network. Tell us about yourself so that we can serve you better.
Sign Up