Association between NCQA Patient-Centered Medical Home Recognition for Primary Care Practices and Quality of Care for Children with Disabilities and Special Health Care Needs. METHODS



This study evaluated the association between NCQA 2008 Physician Practice Connections®-Patient-Centered Medical Home (PPC-PCMHTM) recognition and health service use among Medicaid-enrolled CSHCN in 2010 using multiple comparison group analyses in three states.

Data Sources

This study uses primary and secondary data from multiple sources. We purchased primary data on practices and providers who received NCQA 2008 PCMH-recognition between November 2008 and October 2011 from NCQA. These files contained information on certification level, date of certification, and national provider identifier (NPI) and primary specialty for individual providers within each practice. We obtained secondary data from state Medicaid Analytic eXtract (MAX) 2008 and 2010 eligibility and claims files from the Centers for Medicare and Medicaid Services (CMS). These files contained data on our study populations, including demographics, diagnoses, and health care utilization. State MAX Provider Characteristics (MAXPC) files, also obtained from CMS, contained Medicaid provider identification numbers and NPIs that allowed us to link NPIs from NCQA data file to MAX claims data. We used the American Community Survey (ACS) 2006-2011 public use data file for data on zip code-level sociodemographic characteristics, including poverty, education, employment and languages spoken at home.

State Selection

To be included in this study, states needed to meet the following criteria: (1) relatively high numbers of NCQA-recognized child-serving providers, defined as providers with primary specialty related to pediatrics, family or general medicine; and (2) low penetration of Medicaid comprehensive managed care (CMC) because quality and completeness of MAX claims data for CMC enrollees is suspect. In addition, states had to have MAX 2008 and 2010 data available for analysis to allow for measurement of service use in 2010, and adjustment and matching on baseline service use in 2008 in sensitivity analyses. While seven states met the first two criteria, only three of these -- Louisiana, New Hampshire and Texas -- had 2010 MAX data available. (See Appendix A for additional details on analyses supporting state selection.)

Study Population

The study population included CSHCN age 0-18 years who were enrolled in fee-for-service (FFS) Medicaid in all enrolled months in 2010 and who did not spend more than 90 days in a hospital or long-term care facility. We identified CSHCN using criteria related to: (1) disability status; and (2) diagnoses suggesting a chronic health care need. Children with at least one month of Medicaid eligibility due to disability during 2010 were considered CSHCN. We assumed these children were likely Supplemental Security Income recipients who automatically qualified for Medicaid due to a disability that causes severe functional limitations and can result in death or is expected to last at least one year (Social Security Administration, 2013). In addition, we applied the Chronic Illness and Disability Payment System (CDPS) diagnosis-based software to 2010 MAX claims data (Kronick et al., 2000; Kronick et al., 2009). The CDPS software assigns children to any of 22 different condition categories, and within each condition category, to expected cost categories that may range from "extra high" to "super low" or "not well-defined." For this study, CSHCN included any child flagged in a CDPS condition and cost category, provided that they were not flagged in the pregnancy or low-birth weight categories and were not classified in the "super low," "extra low," or "not well-defined" cost categories within all other condition categories, as these may indicate patients with low complexity of disease and "rule-out" diagnoses, respectively.

Treatment and Comparison Group Assignment

We attributed CSHCN in our sample to the provider in 2010 who supplied the majority of well-child services, other preventive and primary care services, evaluation and management services, and other services that are likely coordinated by a medical home (see Appendix C for a list of diagnosis and procedure codes used for attribution). If there was no majority provider, we attributed children to the provider most recently visited. Over 90 percent and 80 percent of CSHCN in Louisiana and Texas, respectively, were attributed to a provider using this method. The treatment group was comprised of CSHCN attributed to the 114, 145 and 73 providers who received NCQA PCMH-recognition between 2008 and 2010 in Louisiana, New Hampshire and Texas, respectively (N=9,761 in Louisiana, N=4,090 in New Hampshire and N=1,174 in Texas).

We then constructed multiple comparison groups. The first group was a non-matched, "late recognition" comparison group comprised of CSHCN attributed to the 27, 54 and 100 providers who received NCQA PCMH-recognition between January and October 2011 in Louisiana, New Hampshire and Texas, respectively. The rationale for this comparison group was to include children cared for by providers who lagged the treatment group providers in being recognized for meeting NCQA requirements for being a PCMH, but who may have been similarly motivated to obtain it and may be similar to treatment group providers on unobservable characteristics. In addition, we constructed a matched comparison group from Medicaid-covered CSHCN within each of the three states who were not attributed to NCQA-recognized providers. For these matched comparison groups, we conducted exact-matching on age (in years), sex, number of months enrolled in Medicaid in 2010, and disability status, CDPS condition categories, and CDPS prescription drug categories in 2010. The CDPS prescription drug algorithm assigns children to any of 45 different drug categories based on national drug codes from prescription drug claims. We excluded 16 categories from our matching algorithm that either primarily affect the elderly, such as Alzheimer's or osteoporosis/Paget's, or that do not necessarily indicate special needs, such as drug categories for prenatal care, folate deficiency, gastric acid disorder, and infections. Prescriptions filled in any of the remaining 29 drug categories were included in our matching algorithm. We took all available exact-matches within strata, and weighted the comparison children in each stratum to reflect the number of treatment children. For example if three comparison children matched to one treatment child in one stratum, each comparison child received a weight of one-third. We matched 8,414, 3,023 and 968 treatment children in Louisiana, New Hampshire and Texas, respectively, comprising 75-85 percent of the treatment group children, to at least one comparison child each. These comparison children were not linked to particular providers.

Outcome Measures

This study used seven claims-based measures of service use and two claims-based measures of care coordination derived from the initial set of Children's Health Insurance Program Reauthorization Act (CHIPRA) core measures, National Quality Forum-endorsed measures, and widely used Healthcare Effectiveness Data and Information Set (HEDIS) measures. Appendix D provides detailed descriptions of the measure specifications. The following five measures related to service use: any well-child visit, any ED use, any preventable or avoidable ED use (NYU Wagner, 2013), any hospitalizations, and any ambulatory care-sensitive hospitalizations (AHRQ, 2012). Care coordination was measured based on follow-up within 30 days of an ED visit and follow-up within 30 days of a hospitalization. All outcomes were measured in 2010.

Control Variables

Control variables comprised the same set of demographic, Medicaid enrollment, and health status variables used in exact-matching algorithms described above. However, we categorized age based on ages 0-1, 2-5, 6-12 and 13-18 years, as preliminary analyses suggested better model fit with categorical age variables. Because the reliability of race and ethnicity data are unknown in MAX (Mathematica Policy Research, 2011) and there are few other variables on the MAX files related to socioeconomic status (SES) characteristics, we developed proxy measures of SES using zip code-level data from the ACS. These included measures of zip code-level race and ethnicity (percent Hispanic/Latino, percent non-Hispanic/Latino Black, percent non-Hispanic/Latino White, and all other), percent of individuals living in poverty, education levels among women aged 25 and older (percent with less than high school degree, high school degree, some college or college graduate and higher), and employment (percent of adults working full-time versus part-time or not at all).

Statistical Analysis

Our analytic samples included all "late recognition" and matched treatment and comparison children described above with non-missing zip code-level data from the ACS (less than 1 percent of CSHCN in both states had missing zip code data). To test whether CSHCN attributed to NCQA-recognized providers had different patterns of health care utilization than children in the "late recognition," non-matched comparison group, we fit logistic regression models for all our outcome measures by state, adjusting standard errors to account for clustering of children among providers. The only difference for the matched comparison group analyses was to fit weighted logistic regression models to account for multiple comparison children per treatment child. We assessed the magnitude and direction of the coefficients on treatment status, adjusted for control variables listed above, across both "late recognition" and matched comparison group models to assess the strength and consistency of the relationship between NCQA-recognition and the outcome variables. We fit separate models by state due to varying Medicaid programs and policies that may affect provider participation and beneficiary eligibility and enrollment.

To test the robustness of our findings, we tested our models on several subgroups. In matched analyses, we first limited the matched pairs to children residing in the same county to test whether our results are sensitive to treatment-comparison area differences in market area factors that could affect utilization and outcomes. Second, in both matched and "late recognition" analyses, we limited the samples to children ages 2-18 years who were enrolled in 2008 and 2010. The rationale for this subgroup was that by the follow-up year (2010), these children and their parents will be more receptive to changes in providers' practice patterns to improve health care delivery because they will have had more time to develop a relationship with their primary care provider; for the matched analyses, we revised our matching algorithm to include any well-child visits or any ED use in 2008 for this sub-group, and in both matched and "late recognition" sub-groups, we adjusted for any well-child visits, any ED visits and any hospitalizations in 2008 in our regression models. Finally, in the matched analyses, we assessed outcomes among the sub-group residing in the same county in 2010 and who were enrolled in Medicaid in 2008, matching on and adjusting for 2008 utilization as described above.

View full report


"ChildDisV3.pdf" (pdf, 724.89Kb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®