Usefulness of the SF-36 Health Survey in screening for depressive and anxiety disorders in rheumatoid arthritis

Rheumatoid arthritis (RA) is a chronic, painful, progressive condition, which has a substantial impact on patients’ quality-of-life (QoL) [1]. The prevalence of depression in this condition is high, with a recent meta-analysis [2] revealing that an estimated 38.8 % of patients screen positive for probable major depressive disorder (pMDD) according to the 9-item Patient Health Questionnaire (PHQ9; [3]). Common mental disorders such as pMDD or probable generalised anxiety disorder (pGAD) can have implications for long-term health outcomes; depression and anxiety are associated with increased fatigue [4], impaired long-term disease activity and physical disability [5], and reduced treatment efficacy [6].

Despite its prevalence and importance, mental health is rarely measured either in rheumatological research or in clinical practice, reported as an outcome in less than 8 % of published research [7]. QoL is more frequently measured (in 19 % of studies), most often with the Short-Form Health Survey (SF-36 [8]) [7]. The SF-36 has been extensively validated as a measure of QoL in multiple populations and is the most widely used and evaluated QoL outcome measure [9]. The SF-36 consists of 8 domains, which assess physical function (PF), role physical (RP), bodily pain (BP), global health (GH), vitality (VI), social function (SF), role emotional (RE) and mental health (MH). Scores on these subscales can also be combined to create two higher-order summary scores: the physical component summary (PCS) and mental component summary (MCS). The PCS is calculated by positively weighting the 4 physical subscales (PF, RP, BP and GH), and by negatively weighting the psychological subscales (VI, SF, RE and MH). Conversely, the MCS is created by positively weighting the psychological subscales and negatively weighting the physical subscales.

There are several similarities between the SF-36 MH subscale and typical depression and anxiety screening questionnaire. Items relating to low mood (“Have you felt downhearted and low?”), tiredness (“Did you feel tired?”), nerves (“Have you been a very nervous person”) and restlessness (“Did you have a lot of energy”) are comparable to items such as “Feeling down, depressed or hopeless” (PHQ9 item 2), “Feeling tired or having little energy” (PHQ9 item 4), “Feeling nervous, anxious or on edge” (GAD7 item 1) and “Being so restless that it is hard to keep still” (GAD7 item 5). Additionally, the weighting of other QoL domains introduced when combining subscale scores for the MCS include other depression and anxiety symptoms, such as psychosomatic symptomatology and emotional interference with daily activities.

Validating the MH and MCS constructs within the SF-36 as screening tools for depression and anxiety may add extra utility to a questionnaire which is already frequently used for research purposes, and could also provide additional room for interrogation in clinical trial datasets which measure QoL but not mental health. The identification of useful thresholds can also have implications for implementing change in clinical practice. For example, patients attending general outpatient appointment at King’s College Hospital NHS Foundation Trust are required to complete the PHQ9 and GAD7 along with other patient reported outcomes, such as pain and fatigue visual analogue scales (VAS), and the Health Assessment Questionnaire (HAQ [10]) [11], on tablet devices while they wait for their appointment. The results of these assessments are made available immediately on their electronic health record, with advice for onward referral if required [11]. To date, the SF-36 MH and MCS scores have been validated as screening tools for depression and anxiety in an elderly population [12], however a similar validation process has yet to be performed in an RA sample.

We aimed to: 1) examine the relationships between MH and MCS SF-36 domains and depression, anxiety, and indicators of disease severity; 2) assess the accuracy of the MH and MCS SF-36 domains in identifying the presence of pMDD and pGAD, and describe the sensitivity and specificity of cut-off scores in the SF-36 MH and MCS in screening for psychological disorder; and 3) to recommend the most appropriate threshold with which to identify pMDD, pGAD, or presence of any psychological disorder (pMDD or pGAD).

A total of 244 individual patients attended appointments throughout the recruitment period. Of these, 119 (48.8 %) met all eligibility criteria and were invited to participate. Of the patients who did not meet eligibility criteria (N = 28): 2 had severe learning disabilities and could not provide informed consent; 6 patients were too disabled to answer independently; and 10 patients did not speak enough English and did not have a translator present. Nine eligible patients (7.6 %) were unable to be approached due to time constraints in the clinic. Ten eligible patients (8.4 %) declined to participate. A total of 107 patients attending appointments did not have clinic data collected within the previously specified eligible timeframe (same day for psychological variables, ± 3 months for disease activity and disability measures), therefore precluding them from meaningful comparison with the SF-36 data (as SF-36 data would not represent current mood state, or recent disease status). In total, 100 patients were successfully recruited, yielding a total participation rate of 84.0 % (Fig. 1).

Table 1 shows the mean scores of pertinent available mental and physical variables across all patients attending appointments during the recruitment period. In comparison to patients who were recruited, ineligible patients were significantly older (p = 0.007), and had higher levels of pain (p = 0.03), disability (p = 0.01), and DAS28 (p = 0.02). Neither depression nor anxiety levels significantly differed across participation statuses.

The results of this analysis suggest that the SF-36 can be used to determine the presence of pMDD, pGAD, or general psychological disorder, and potential thresholds have been suggested for these diagnoses. However optimal use of the SF-36 for screening for mental disorder may be by utilising a threshold of ≤38 on the MCS, to identify the presence of any psychological disorder. This threshold had good sensitivity (88 %) and specificity (80 %), and correctly classified 83 % of patients with either pMDD or pGAD. This can be compared to sensitivity and specificity of 95 and 66 % respectively for the NICE-recommended questions for identifying depression in patients with chronic physical health problems [13].

A similar validation study in an elderly population identified 42 as an appropriate MCS threshold to identify significant psychological distress [12]. Our identified threshold of 38 is comparable to this. As the MCS is normed, with a score of 50 (standard deviation = 10) representing the UK population, this threshold identifies patients with a score of 1 standard deviation below the population mean (scoring in the bottom 16 % of the population) as being at risk of pMDD or pGAD.

There are several elements to take into consideration when evaluating the research process described. The primary limitation is the lack of a “gold-standard” depression measure with which to validate the SF-36 domains. Employing a psychologist or psychiatrist to diagnose the presence of depression was beyond the financial and time restrictions present. We therefore relied on the measurement already taken as part of routine clinical practice, which were the PHQ9 and GAD7. Given their sensitivity and specificity of these measures against gold standard diagnostic criteria, we can expect an increase in detection of false positives. For example, if the true prevalence of depression is 20 %, and assuming an accuracy of 89 % for the PHQ9 [14] and GAD [15], our detected MCS accuracy for pMDD of 83 % would translate into a lower bound for the accuracy of 74 % for detecting true MDD. This assumed the two screening tools are independent, so it is almost certain that true accuracy is between 74 and 83 %. Although the PHQ9 and GAD7 are well-validated themselves [14, 15] future research is required to replicate the findings of this study using a gold-standard psychiatric interview against which to validate the MCS and MH scales of the SF-36. However this research has increased the utility of this widely used questionnaire and provides interesting scope for further validation and secondary analyses of existing datasets.

An additional consideration is the representativeness of our sample. As described in our methods section, patients who were unable to give consent due to learning disabilities or dementia, or who were too disabled to answer independently, we not considered eligible for inclusion in this study. This selection bias resulted in ineligible patients being significantly older, having higher levels of pain, disability and disease activity, than patients who were recruited for participation. The generalisability of our results to the wider RA population is questionable, although it is important to note that there were no differences in mental health across those who participated and those who did not.

Despite these shortcomings, this analysis provides a useful first step in adding extra utility to a measurement tool already frequently used for research purposes. The addition of regular HRQoL measurement to routine RA clinical care would be beneficial, as QoL is a key outcome of importance to patients [21], and doing so would also support the need for embedded mental health care within rheumatological management practice.