Depression is a public health problem. In 2000, it was rated as the fourth leading cause of disease globally, accounting for 4.4% of total disability adjusted life years (1-3). Depression affects approximately 14.8 million American adults, or 6.7% of the population (4), and is linked to risk behaviors such as smoking, alcohol use, and physical inactivity (5-7). It is also associated with chronic diseases such as diabetes, asthma, arthritis, cardiovascular disorders, and cancer (2,8). Depression is underdiagnosed and undertreated (1,2,6). Getting a reliable estimate of depression prevalence in the general population is important but presents challenges. Several measures exist for detecting depression in clinical settings (9), and multiple self-reported questions used for clinical diagnosis have been adapted for use on population-based surveys (6,10). The challenge is finding a simple means to estimate depression prevalence in the general population.

The Behavioral Risk Factor Surveillance System (BRFSS) is a telephone survey administered in all 50 states, the District of Columbia, Puerto Rico, the US Virgin Islands, and Guam with funding and specifications from the Centers for Disease Control and Prevention (CDC) (11). The BRFSS monitors the prevalence of behavioral risks for the leading causes of disease and death among adults in the United States (11). A 9-question health-related quality of life (HRQOL) module has been available since 1995 (12). In 2006, a 10-question depression and anxiety (D&A) module was also made available to states. Rhode Island was the only state to include both modules on its 2006 BRFSS questionnaire.

We compared depression and mental distress estimates from the HRQOL and D&A modules using data from Rhode Island's 2006 BRFSS. Two of the 5 items that are related to mental health in the HRQOL module were used, 1 for sad/blue/depressed and 1 for frequent mental distress. Two measures from the D&A module were used. The D&A module includes the patient health questionnaire 8 (PHQ-8), which is used to create a 5-point scale for depression severity based on an algorithm using responses to 8 questions. Severity scores can be grouped to create a dichotomous variable for current depression (6,9). The first 2 questions of the PHQ-8, called the PHQ-2, is also used to provide a simple measure of current depression (9). Our hypothesis was that either of the HRQOL questions or the PHQ-2 can serve as a proxy for the PHQ-8 on the BRFSS. The objective of this study was to assess whether 1 or 2 questions on depression and mental distress can yield prevalence estimates of depression comparable to those from the PHQ-8, which has a high degree of sensitivity and specificity.

Compared with the PHQ-8, each of the 3 indicators had a high level of specificity, ranging from 94.4% to 96.4%; the PHQ-2 had a slightly higher negative predictive value than the other 2 indicators (Table 1). The sensitivity of the 3 indicators was weaker, ranging from 59.4% to 66.8%. The PHQ-2 had higher sensitivity than HRQOL indicators. The positive predictive values for the PHQ-2 and the sad/blue/depressed indicator were almost identical; the "frequent mental distress" indicator had a lower positive predictive value.

In the HRQOL module, the prevalence of frequent mental distress among Rhode Island adults was 10.3% and of sad/blue/depressed was 8.9% (Table 2). In the PHQ-2 and PHQ-8, the prevalence of current depression was 9.8% and 8.6%, respectively. Results of tests for significance for sad/blue/depressed and for PHQ-8 were consistent for the 11 demographic and risk variables with the exception of results for age. Results of tests for significance for PHQ-2 and for PHQ-8 were consistent for the 11 demographic and risk variables with the exception of results for age and sex. Results of tests for significance for frequent mental distress and for PHQ-8 were consistent for all of the 11 demographic and risk variables. Prevalence estimates for the 11 demographic and risk variables for each of the indicators were comparable with one another.

We used the area under receiver operating curves (AUC) to assess model discrimination. The AUC for frequent mental distress, sad/blue/depressed, PHQ-2, and PHQ-8 were 0.74, 0.80, 0.78, and 0.83, respectively. The AORs for sad/blue/depressed and for PHQ-8 (Table 3) were consistent with one another for 6 predictors. Both indicate increased odds of depression for women, annual income less than $25,000, unable to work, current smoker, no leisure-time PA, and disability. The AORs for PHQ-2 and for PHQ-8 are consistent with one another for 6 predictors. Both indicate increased odds of depression for younger adults, annual income less than $25,000, unable to work, current smoker, no leisure time PA, and disability. The AORs for frequent mental distress and for PHQ-8 are consistent with one another for 8 predictors. Both indicate increased odds of depression for adults aged 18 to 44 years and 45 to 64 years, annual income less than $25,000, unable to work, current smoker, asthma, obesity, and disability.

The prevalence of sad/blue/depressed in Providence (excluding the affluent east side), West Warwick, and Warwick ranged from 11.7% to 13.6%, higher than the rest of the state (Figure 1). The prevalence in Woonsocket, Central Falls, Pawtucket, North Providence, Johnston, Rumford, East Providence, Cranston, and Riverside ranged from 7.3% to 11.6%. These areas with higher depression rates include the more urban areas of the state, which have a higher proportion of low-income households than do the suburban and rural areas.

The prevalence of current depression in Providence (excluding the affluent east side), Central Falls, Pawtucket, Warwick, and West Warwick ranged from 11.3% to 15.5% (Figure 2). The first 3 cities are urban with a high proportion of low-income and minority residents. The prevalence in Woonsocket, East Providence, Coventry, Greene, West Greenwich, and East Greenwich ranged from 7.8% to 11.2%. Other than East Providence, these are suburban areas. The remainder of the state, with the lowest rates for current depression, is largely suburban and rural.

Our analysis showed that any of the 3 shorter items provide results comparable with those of the PHQ-8 in estimating overall prevalence of depression and mental distress, identifying high-risk populations, and identifying significant associations with risk variables. We recommend use of any of the 3 shorter items as a proxy on the BRFSS or similar population-based surveys to obtain a population estimate of depression prevalence. Any one of the 3 is adequate for use in descriptive analyses of population data. They provide an efficient means of assessing depression prevalence in adult populations when survey efficiency precludes use of the longer PHQ-8. However, the PHQ-8 should be preferred for surveys requiring a high level of sensitivity as well as specificity, accuracy in reliably assigning depression severity status to specific respondents, or requiring assessment over time of population changes in depression severity.

The strengths of the PHQ-8 are its high degree of both sensitivity (88%) and specificity (88%) for major depression (6,9), its adequacy for diagnosis, its ability to assess severity of depression, and its validation in the general population (6,9,10). The weakness of the PHQ-8 is its length, which makes it time-consuming to administer and answer, and its complex scoring algorithm. Although the simple question measures of depression are inadequate for diagnosis (9), they are useful for screening for depression. Sad/blue/depressed and frequent mental distress are single questions, very easy to answer and administer, with few training requirements. The PHQ-2 is easy to answer and administer, and the scoring algorithm is simple (9).

The test characteristics and the test performance of the PHQ-2 were more sensitive than the 2 HRQOL indicators. Frequent mental distress was more concordant with the PHQ-8 than was the PHQ-2 in distinguishing demographic groups and risk factors with higher prevalence of depression. Based on the AUC, sad/blue/depressed is better than frequent mental distress and the PHQ-2. The proportion of records with missing values was lower for the 2 single HRQOL items than for the PHQ-2 or the PHQ-8. No measure was better than the others in all respects.

Both the sad/blue/depressed and the PHQ-8 maps showed that the prevalence of depression was highest in the core urban areas of the state and lowest in the more suburban and rural areas. The differences between the 2 distributions may reflect the greater power of the PHQ-8 to discriminate between cities and towns. This discriminatory power may also be reflected in the wider range of values resulting from the PHQ-8 measure than from the sad/blue/depressed measure.

Rhode Island is a small state, so the study population is homogeneous compared with populations of other states. Our analyses went beyond simple comparisons of test characteristics and test performance to include distinguishing levels of demographic characteristics and risk factors as well as within-state geographic comparisons.

Some limitations of this analysis should be noted. First, the HRQOL indicators are based on a 30-day recall period, while the PHQ-2 and PHQ-8 are based on a 14-day recall period. We have no way to assess the effect of this difference in recall periods. Second, both sets of questions (the HRQOL and the PHQ-8) were asked in the same interview session (ie, they were not context-independent). We have no way to assess the effect of this. Finally, because the calculation of PHQ-8 and PHQ-2 scores require responses to all questions used in calculating scores, it was necessary to impute missing values. In the future, we need to vary the cut points of the 2 single-question screeners and the PHQ-2 to optimize their performance against the PHQ-8 (30).

We conclude that, for Rhode Island, any of the 3 short screeners is sufficiently specific and sensitive to provide population prevalence estimates of depression and can be used for descriptive analyses of our population survey data. Mapping of these depression estimates also indicates localities where the need for mental health services is greatest. To validate the generalizability of our findings, it will be important to replicate them in other states' BRFSS surveys.