Do Secondary Conditions Explain the Relationship Between Depression and Health Care Cost in Women With Physical Disabilities?

Initial interview 

The enrollment interview included questions about basic demographic information, current disabilities, and health status. Information was gathered about disability including disability type, age at onset, and duration. To identify disabilities, participants chose from a list of 16 physically disabling conditions plus an open-ended “other condition” category. If participants endorsed more than 1 of these conditions, they were asked to indicate which one was the most limiting; this condition was considered their primary disability.15 Participants also completed the BDI-II16 and HCC.5, 17, 18 The HCC includes 42 secondary and chronic health conditions and constitutes an adaptation of the Secondary Conditions Screening Instrument.17 Like the study protocol used by Nosek et al,5 we excluded conditions that resulted from, or reflected participants' interaction with, their environment, such as difficulties with access and equipment injuries. For our study purposes, we added female-specific conditions including yeast infection, vaginal infection, and menstrual problems.

Bimonthly interviews 

During each of the 6 bimonthly interviews, participants completed the BDI-II, HCC, and a health care utilization questionnaire developed for this study and based on the Health and Social Service Utilization Questionnaire19 and the Stanford Health Assessment Questionnaire.20 Our questionnaire collected information on inpatient, emergency department, and outpatient utilization during the past 2 months. Information on inpatient utilization included length of stay, primary and secondary diagnoses, primary reason for visit (ie, primary disability, depression, or other), and placement at discharge (eg, discharged to home). Information on emergency department visits included primary reason for visit. Information on outpatient visits included primary reason for visit and type of physician seen (ie, primary care physician, specialist, mental health care provider, other).

Missing visits and data aggregation 

We assessed health care use every 2 months to reduce recall bias on the part of the study participants, and we aggregated reported use for each person across the 1-year follow-up period. We excluded from analyses all participants who missed more than 1 interview. For women who missed 1 interview, we imputed the missing values for outpatient visits using the mean number of outpatient visits reported during their other 5 interviews. Emergency department visits and inpatient stays were much less frequent; therefore, we used a regression-based method to impute emergency department visits and inpatient stays for participants who missed 1 interview. To impute inpatient visits, we regressed the number of inpatient stays onto the total number of outpatient visits during the entire study period using data from women who had not missed any interviews. Using the intercept and outpatient visit coefficients from this regression, we then calculated the expected number of inpatient stays for the missing interview period. We used the same procedure to estimate emergency department visits during the lost 2-month interview period.

Assignment of costs 

We measured direct costs by assessing the patient's use of health care services.21, 22 We identified all medical resource use as described above and then identified a dollar cost for these units based on Medicare reimbursements. For inpatient stays, a certified medical coder assigned a probable International Classification of Diseases–Ninth Revision–Clinical Modification diagnosis code based on the participant's reported primary reason for hospitalization, and we computed direct costs according to the average national price for the appropriate diagnosis related group. Outpatient costs were assigned by our medical coder using the Berenson-Eggers type of service coding system (http://www.cms.hhs.gov/hcpcsreleasecodesets/20_betos.asp). The Berenson-Eggers type of service codes consist of readily understood clinical categories that permit objective assignment. Outpatient costs were determined using the most recent cost data available (2002) at the time of the study for each Berenson-Eggers type of service code. The investigators, study staff, and the study's medical coder jointly reviewed all inpatient and outpatient coding. Total costs were computed as the sum of inpatient and outpatient costs.

Depressive symptomatology 

We categorized participants as ever depressed if they scored 17 or higher on the BDI-II16 at any of the interviews. We further categorized participants into 1 of 4 categories based on the number of interviews in which their BDI-II scores were 17 or higher: never depressed (BDI score <17 at all interviews), depressed some of the time (BDI score ≥17 at 1–3 interviews), depressed most of the time (BDI score ≥17 at 4–6 interviews), or always depressed (BDI score ≥17 at all interviews). Participants who were missing 1 or more BDI scores were categorized based on the percentage of times their BDI scores were 17 or higher. Finally, we calculated an average of all recorded BDI scores over the full year for each participant.

Other secondary health conditions 

Respondents were asked on the HCC to rate the severity of 42 secondary and chronic health conditions on a scale of 0 to 4 (0, not a problem in the previous 2 months; 1, mild or infrequent problem; 2, moderate or occasional problem; 3, significant or chronic problem; 4, never had this condition). We classified participants as endorsing a secondary condition if they responded with 1, 2, or 3. Participants responding with 0 or 4 were considered to not have that secondary condition. The interference ratings for each secondary condition endorsed by the participant were summed to assess the cumulative impact.5 Although depression is included on the HCC, we specifically excluded it from the HCC severity score. As a result, the HCC severity scale could range from 0 to 123, with higher scores indicating greater interference from secondary conditions. For the purposes of these analyses, HCC severity scores were averaged across all interviews.

Descriptive analyses 

We compared demographic and disability characteristics of participants who were included in these analyses with participants who were excluded (>1 missed visit). In addition, we compared the demographic characteristics of participants with complete data with participants who missed 1 interview. We estimated the occurrence of depressive symptoms during the study period across the entire sample and within specific disability groups.

Analysis of health care utilization and costs 

We compared the average numbers of outpatient visits, emergency department visits, and inpatient stays between ever depressed and never depressed groups of women. Ever depressed women were further subcategorized by the frequency (ie, depressed some of the time, most of the time, all of the time) and the severity (average BDI-II score) of their depression.

We then ran 4 sets of linear regression models on our aggregated total direct cost measure to determine which factors affected participant costs. For the first set, the primary explanatory variables were dummy variables for primary disability category and presence of depressive symptoms at any point during the study (ever depressed vs never depressed). We included age, race, income level (categorized as: ≤$15,000; >$15,000; not reported), education, and length of disability as covariates. In the second set of regression models, we replaced the dichotomous depression variable with categorical variables for the 4 frequency categories of depression. In the third set of regressions, we used the year-long average BDI score. Within each of these 3 sets, we ran the regression model with and without the average HCC severity score (omitting depression) as a predictor to test whether the cumulative severity of secondary health conditions affected any of the relationships observed between depressive symptomatology and total cost of care. We then ran a final regression model to test the association of the average HCC severity score with total cost in the absence of our measures of depressive symptoms.