The timeframe of this project, 1994–1999, corresponds to the period of a “natural experiment” during which there was a substantial increase in the use of managed care in the Medi-Cal program. Prior to 1994, five of California's 58 counties participated in a demonstration project of mandatory Medi-Cal managed care. During the period of the study, 16 additional large urban counties in the state (where approximately 85 percent of the Medi-Cal beneficiaries in the state reside) also implemented a mandatory managed care program for all of their TANF-eligible Medi-Cal beneficiaries. The implementation of Medi-Cal managed care took somewhat different forms in the affected counties. The main difference was whether the managed care was provided by a county (public) operated health plan, commercial (private) health plans, or some combination of the two. Prior to this mandated requirement, TANF-eligible Medi-Cal beneficiaries may have either been in fee-for-service or voluntary managed care, and counties that did not implement mandatory Medi-Cal managed care during the study period continued to have Medi-Cal beneficiaries in either fee-for-service or voluntary managed care.
The data for this study come from two main sources: (1) the annual California hospital discharge data available from the California Office of Statewide Health Planning and Development (OSHPD), and (2) the California Department of Health Services (DHS) Medi-Cal monthly eligibility file. The California hospital discharge record includes among other things, information on admission month and year, patient demographics, and diagnosis and procedure codes. OSHPD applies several hundred audit rules to ensure the validity of their data before making them available and would not release the key data elements used in this study if they exceeded an error tolerance level of 0.1 percent (Office of Statewide Health Planning and Development 2000
). This file also contains a field indicating expected source of payment, but there are reasons to suspect that this information may be inaccurate, particularly for Medi-Cal beneficiaries. To enhance the accuracy of whether a hospitalized individual was in fact a Medi-Cal beneficiary, we used a special research file that linked hospital discharge data with the DHS Medi-Cal eligibility file for the period of 1994–1999. This file provided additional information for the entire year on patients' month-by-month Medi-Cal enrollment status, aid category, health plan, and the county of residence. These data elements, combined with DHS supplied information on the date in which a California county implemented mandatory Medi-Cal managed care, enabled us to classify each hospitalization as occurring for a Medi-Cal beneficiary in fee-for-service, voluntary managed care, or mandatory managed care. Overall, 93 percent of Medi-Cal hospitalizations recorded by DHS could be linked using patient identifiers, such as social security numbers, with a hospitalization record in the OSHPD file. The 7 percent that could not be matched were mostly among newborns whose records could not be distinguished from their mothers' (Rains and Tagupa 2001
Since we are using hospitalizations as an indicator of ambulatory care prior to the hospitalization, we chose to assign to Medi-Cal only those hospitalizations in which an individual had Medi-Cal coverage in the month prior to hospitalization. In this way we avoided the misclassification of an uninsured individual who gained Medi-Cal as a result of their hospitalization. However, this approach required that we exclude January admissions from our analysis as information on an individual's Medi-Cal eligibility was only available for the calendar year, and we could not determine if someone with a January admission was a Medi-Cal beneficiary in the previous December. Also since our hospitalization discharge and enrollment files were linked to a calendar year, we could not accurately calculate admission rates for hospital admissions that resulted in discharges in a different calendar year. Less than 1 percent of admissions had discharges in a subsequent year and these were excluded from the analysis.
We captured Medi-Cal beneficiaries' hospitalizations in states that border California by searching for Medicaid beneficiaries with California zip codes in Arizona's, Nevada's, and Oregon's hospital discharge abstracts records for the same time period. Ambulatory care sensitive condition hospitalizations of California residents in these three states totaled less than 0.2 percent of such hospitalizations within California.
Data on number, demographics, eligibility category, and health plan type of the entire Medi-Cal population (not just those hospitalized) were obtained from DHS's Medi-Cal monthly eligibility file. The enrollment files for years prior to 1996 contained information only as of the first month of each quarter (January, April, July, and October). We used linear interpolation to obtain the estimate for the other 8 months of those years.
It was our hypothesis that, if Medi-Cal managed care was having a positive effect on Medi-Cal beneficiaries' access to ambulatory care, then we would observe a lower rate of ambulatory care sensitive admissions among Medi-Cal beneficiaries in managed care than we would among Medi-Cal beneficiaries in fee-for-service care. Assuming that voluntary Medi-Cal managed care would disproportionately attract healthier Medi-Cal beneficiaries than would be assigned under a mandatory Medi-Cal managed care program, we also expected to observe lower rates of admissions for ambulatory care sensitive conditions in voluntary than mandatory Medi-Cal managed care.
If mandatory Medi-Cal managed care completely eliminated the opportunity for beneficiaries to choose between managed care and fee-for-service, then the difference in hospitalization rates for ambulatory care sensitive conditions between voluntary and mandatory managed care could be assumed to be because of health selection bias in voluntary managed care. However, even in counties where enrollment in managed care was mandatory, TANF-eligible Medi-Cal beneficiaries were given an opportunity to apply to remain within fee-for-service care. In 1999 approximately 10 percent of TANF-eligible Medi-Cal beneficiaries in mandatory managed care counties were still in fee-for-service care. We suspected that it was the sicker patients with higher hospitalization rates who would have been most likely to opt out of mandatory managed care. To address this potential bias, we performed an “intention to treat” reanalysis to arrive at a more conservative estimate of the effect of mandatory Medi-Cal managed care. In this analysis, beneficiaries were categorized as fee-for-service or mandatory managed care not on the basis of their actual Medi-Cal health plan assignment, but by whether their intended assignment was to be in mandatory managed care on the basis of their aid category, county of residence, and a date of admission after the implementation of mandatory Medi-Cal managed care in their county.
We used commonly accepted lists of conditions defined with diagnostic codes for children and adults to calculate the number of hospitalizations for ambulatory care sensitive conditions for Medi-Cal beneficiaries (see online-only Appendix available at http://www.blackwell-synergy.com
) (Millman 1993
; Billings et al. 1993
). Hospitalizations were considered to be for ambulatory care sensitive conditions when any of the ICD-9 codes for these conditions were listed as the primary reason for the admission. We also counted the number of Medi-Cal hospitalizations for the nonambulatory care sensitive condition of appendicitis under the assumption that if the mechanism through which Medi-Cal managed care lowered rates of ambulatory care sensitive conditions was improving access to ambulatory care, we should not observe differences between Medi-Cal managed care and fee-for-service in the admission rates for appendicitis. Appendicitis is a nonambulatory care sensitive condition because there are no ambulatory care based strategies for preventing a hospitalization for this condition. However, appendicitis admission rates could still vary in association with the Medi-Cal delivery model as physicians exercise discretion in determining whether or not to admit and operate on patients who present with symptoms that may be consistent with appendicitis knowing that some of those cases will not ultimately have appendicitis. If we found lower rates of admission for ambulatory care sensitive conditions and the nonambulatory care sensitive condition, appendicitis, in Medi-Cal managed care this would imply that managed care was associated with a higher threshold for admission rather than better access to ambulatory care.
For the duration of the entire study period (1994–1999), we calculated the average monthly rate of hospitalization for ambulatory care sensitive conditions and for appendicitis for TANF-eligible Medi-Cal beneficiaries in fee-for-service, voluntary managed care, and mandatory managed care. We limited the analysis to individuals who were less than age 65 under the assumption that older individuals were likely to also have Medicare insurance. The numerator of the rate was the count of hospitalizations for the specified conditions in a given month. Using a record linkage number that enabled us to determine whether the admission was the first for an individual in our dataset or a readmission, we calculated separate rates counting the number of beneficiaries with at least one hospitalization for an ambulatory care sensitive condition and then counting all of their hospitalizations for these conditions. We did this for two reasons: (1) we wanted to assess whether managed care might be exerting its effect on initial admissions that are more related to access versus readmissions that are more related to the quality of care for individuals already recognized as having an ambulatory care sensitive condition, and (2) we thought it was possible that some unobserved factors may have predisposed some individuals with multiple admissions to be more likely to be in fee-for-service versus managed care. The denominator population for calculating the admission rate for each Medi-Cal delivery model was obtained from the Medi-Cal monthly eligibility file.
Recognizing that nonrandomly distributed patient and county characteristics could confound our results, we used multivariate Poisson; regression analysis to model the monthly ambulatory care sensitive condition admission rate as a function of the Medi-Cal delivery model (fee-for-service, voluntary managed care, and mandatory managed care) controlling for admission month, admission year, patient age (0–17 versus 18–64 years), sex, race/ethnicity (African-American, Asian and Pacific Islander, Hispanic, non-Hispanic White, and Other), and county of residence. Adjusting for month controlled for seasonal variation in admission rates while including year of admission in the model controlled for secular changes in the hospitalization rate for the study conditions. The inclusion of county of residence in the model adjusted for unmeasured differences across counties in hospitalization rates for ambulatory care sensitive. Furthermore, the county of residence variable accounted for within county clustering of these rates over time. In order to account for any residual clustering or correlation, we also corrected for any remaining overdispersion in our model by using a scale factor that was equal to the value of the square root of Pearson χ2
divided by the degrees of freedom (McCullagh and Nelder 1989
). As a further check against the possibility that residual autocorrelation could alter our results, we evaluated a model that included autocorrelation, using the generalized estimating equations method (Liang and Zeger 1986
). We found that the estimated autocorrelation was only 0.08 and our results did not differ in any substantial way (data not shown).
Observations of the number of hospitalizations for ambulatory care-sensitive conditions were ascertained from the hospital discharge files and were aggregated into analytic cells defined by different combinations of values for the independent variables. For example, one cell corresponded to the number of hospitalizations for ambulatory care-sensitive conditions among TANF-eligible Medicaid beneficiaries in fee-for-service care who in February 1994 were between 0 and 17 years of age, female, African-American, and residing in Los Angeles county. Such an approach can accommodate changes in individual characteristics over time, such as type of health plan held by a beneficiary as fee-for-service or mandatory managed care. Using the combination of the values of the specified independent variables, there were 229,680 different possible cells in which to place observations. Since the data from out-of-state admissions did not include information on Medi-Cal eligibility category we were unable to include them in our multivariate model. The corresponding denominator population for calculating the admission rate for each cell was obtained from the Medi-Cal monthly eligibility file that had detailed information on each of our independent variables. The number of Medicaid beneficiaries “at risk” for a hospitalization for an ambulatory care-sensitive condition varied across cells and was included as an offset variable in the model. The coefficient estimates from Poisson regression model were used to obtain predicted rates standardized to adjust for differences in group composition. As with the unadjusted results, we performed separate multivariate analyses using counts of all ambulatory care sensitive condition hospitalizations, and counts of persons who had at least one ambulatory care sensitive condition hospitalizations in the 6-year study period.
To explore whether there was a differential effect of Medi-Cal managed care on ambulatory care sensitive condition admission rates that was dependent on the race or ethnicity of the Medi-Cal beneficiary, we repeated the multivariate analyses including an interaction term for patient's race or ethnicity by Medicaid delivery model (fee-for-service, voluntary managed care, or mandatory managed care).
The study protocol was reviewed and approved by the University of California San Francisco Committee on Human Research and the California Department of Health Services Committee on the Protection of Human Subjects.