Beneficiary Group Definitions

In the Beneficiary Reports, scores are reported for 8 beneficiary groups: Prime enrollees, Enrollees with a Military PCM, Enrollees with a Civilian PCM, Select Enrollees, Purchased Care Users, Retirees and their Dependents, Active Duty, and Active Duty Dependents. These categories are assigned based on a combination of survey responses and DEERS data. Prime enrollees consist of active duty and members of other beneficiary categories who say they use Prime most, who report that they have been enrolled in their current plan for at least 6 months. Enrollees with a military PCM are Prime enrollees who, according to DEERS, are enrolled to a military PCM. Enrollees with a civilian PCM are Prime enrollees who, according to DEERS, are enrolled to a civilian PCM. Select enrollees are non-active duty TRICARE users who say they use TRICARE Select for most of their care. Purchased Care users are those who use TRICARE Select, TRICARE reserve select, or TRICARE retired reserve, or are enrolled with a civilian PCM. Active duty, active duty family members and retirees and family members are identified as such by DEERS from TRICARE users.

Demographic Adjustments

All scores in the TRICARE Beneficiary Reports are adjusted for patient characteristics affecting their scores. The purpose of risk adjustment is to make comparisons of outcomes, either internally or to external benchmarks, that control for characteristics beyond the health care provider's control. Based on previous work with satisfaction scales derived from CAHPS, it appears that satisfaction increases with age and decreases with poor health across social classes and insurance types. Thus, scores are adjusted for age and health status.

The methodology used is an adaptation of that found in CAHPS 2.0 Survey and Reporting Kit (DHHS, 1999). The model used for this adjustment is: 

where Y_ijk is a dependent variable, β_qk's are parameters to be estimated, A_qk's are age dummy variables (A_qlk = 1 if the beneficiary is in age group q, and 0 otherwise; ₁ = age 18-24, A₂ = age 24-34, A₃ = age 35-44, ₄ = age 45-54, A₅ = age 55-64, A₆ = age 65-74, and A₇ = age 75 and older), P_k is health status. The subscripts i, j, and k refer to the beneficiary, region and enrollment group, respectively.

Given 15 regions, the specifications that we use are:

where R_j's are regional dummy variables (R_jk = 1 if the beneficiary is in region j and beneficiary group k, and 0 otherwise).
For beneficiary groupk, the adjusted regional value is:

 where  is weighted proportions of age group q in beneficiary group k.

Standard errors then can be estimated as the standard error of residuals for catchment areas or regions using SUDAAN. These standard errors can be used in hypothesis tests comparing adjusted values to other adjusted values or to external benchmarks. Composite values are calculated as averages of regional or catchment area adjusted values for questions making up the composites, in which each question is equally weighted.

Benchmarks are also adjusted for age and health status as are scores taken from survey responses. If the benchmark data set contains age and health status information, we fit a model of the form:

where the A's are age groups and P is health status. Then the adjusted benchmark is:

using the mean values of A and P for beneficiary group k. 

The adjusted values for that beneficiary group can then be compared to a benchmark appropriate for their age distribution and health status.

In the Beneficiary Reports a single benchmark is presented in comparison to many beneficiary groups. We accomplish this by recentering scores for beneficiary groups. Each score and benchmark is calculated for the appropriate beneficiary group. Then a recentering factor for each beneficiary group is calculated as the difference in adjusted benchmarks between a beneficiary group and the all users group. For the all users group, that recentering factor is zero. The recentering factor is added to score for each region or catchment area for that beneficiary group. Thus beneficiary groups can also be compared controlling for age and health status and can be compared to the same benchmark.

Calculating Scores

Beneficiary Reports include three types of scores: CAHPS composites, ratings, and a preventive care composite. The healthy behaviors composite is calculated like CAHPS composites.

The preventive care composite is calculated as P_i =Σw_ir_i, where w is the proportion of the population eligible for all others for whom a preventive care measure i is relevant and r is the proportion of that eligible group receiving preventive care. For example, if i is mammography, w is the number of women over age 40 over the number eligible for mammography plus the number eligible for cholesterol screening and so on. The value r is the proportion of these women receiving mammography. In order to make the composite comparable between years, regions, and beneficiary groups, the weight, w, is calculated from the sample of TRICARE users who responded to surveys in CY 2011. Both the numerator and denominators for these rates are calculated using survey sampling weights. 

CAHPS composites are calculated as S_i=(1/n_i) Σ(q_j/k _j), where n_i is the number of questions in the composite i, q_j is the (weighted) number giving a favorable response to question j in the composite i, and k_j is the (weighted) number responding to that question j. The values q_i and k_j is calculated using sampling weights. CAHPS ratings are calculated as S _i=q_i/k_i, where q_i is the number giving a favorable response and k_i is the (weighted) number responding to rating i. All scores are adjusted for age and health status (see above).

Standard Error

Standard error (i.e., the square root of the variance) of an estimate represents precision (or sampling error) of the estimate from a particular sample. It indicates how much variability there is in the population of possible estimates of a parameter for a given sample size. In this survey, the standard error is calculated using the Jackknife replication method based on 60 replicate weights. It takes into account the sample design, as well post data collection weighting adjustment process such as non response adjustment, post-stratification, and weight trimming. Please see the Health Care Survey of DoD Beneficiaries: Adult Technical Manual for the construction of the replicate weights; and Wolter, K. M. (2007), Introduction to Variance Estimation. Second Edition. Springer-Verlag. New York, pp. 181-182, for the calculation of standard error.

Testing Trends

In the Beneficiary Reports, we use linear regression to estimate and test for statistical significance a quarterly rate of change. Our estimate for the rate of change is:

where t is the quarter, St is the score and wt is the total weight of quarter t's observations. In order to test the hypothesis that trend is zero, we use the standard error for the trend coefficient:

where σ_t is the standard deviation in quarter t. The hypothesis test is based on a t-test of the hypothesis that T=0, where n is the total number of observations for all 3 quarters p=Prob(abs(T/S)>0,n).