Participants

The study was a face-to-face, cross-sectional survey of a representative sample of adult attendees (18 years or older) at primary care health centres in Catalonia. Catalonia is one of the 17 regions or ‘autonomous communities’ that comprise Spain. As a consequence of a devolution process started in 1981, the autonomous communities have full governance on health and social care. Healthcare and social care for people with severe disabilities are publicly financed and near-universal coverage is provided. The features of this system have been explained elsewhere. ^{Reference Salvador-Carulla, Garrido, McDaid and Haro11}

A stratified multistage probability sample without replacement was drawn. Replacement was prohibited to ensure that every individual had a known probability of selection. The sampling frame was all the health regions in Catalonia (a total of seven). The first stage consisted of selection of the primary care centres within each health region, with the number of primary care centres selected in each region proportional to its population. However, in order to have a minimum set of interviews even in the smaller regions, at least six primary care centres were chosen per region. Each primary care centre's selection probability was related to the population of the catchment area covered by the centre. A total of 77 health centres out of 352 participated in the DASMAP study. In the second stage, all the general practitioners (GPs) at the selected health centres were invited to participate and a total of 618 GPs did so. This represented nearly 69% of all the GPs working at the 77 health centres. The third stage consisted of the random selection of patients. Participants were selected with a systematic sampling strategy from the daily list of all patients with an appointment with each of the participating GPs. A total of 3815 participants were evaluated. The weighted response rate was 80.5%. Further information on the DASMAP study can be found elsewhere. ^{Reference Serrano-Blanco, Palao, Luciano, Pinto-Meza, Luján and Fernández12}

Measures

Procedure

A group of 20 trained psychologists evaluated participants. An expert panel of three trained the psychologists during a 2-day course. Data were collected between October 2005 and March 2006 using a paper-and-pencil personal interview. After an appointment with a GP, individuals were invited to participate in the study. They were evaluated at their primary care centres after acceptance (including provision of signed informed consent). During a clinical interview of approximately 45 min the instruments were administered. After data collection, responses were processed using the response automatic-capture software TeleForm for Windows (Autonomy Cardiff, www.cardiff.com/products/teleform/index.html). Ethical approval was obtained from the Sant Joan de Déu Foundation ethics board.

Statistical analyses

The dependent variable was the SF–6D score utility. As SF–6D scores had skewed distributions, we used a non-parametric approach. Statistical analyses were carried out in four steps. First, we looked at sociodemographic factors (gender, age, marital status, education and employment), mental disorders and chronic physical condition distributions for the participants. Proportions were weighted in order to restore the representative validity of the sample. We also presented the median of the SF–6D by sociodemographic and clinical (mental disorders and chronic physical conditions) variables. The proportion of participants reporting full health and the most frequent health-state profile were also calculated. Comparisons in the SF–6D by sociodemographic and clinical variables were conducted using the Kruskal–Wallis test. Second, to model the association between health conditions (mental disorders and chronic physical conditions) and the SF–6D, we used non-parametric quantile regression. Quantile regressions extends beyond the notion of ordinary least squares, which estimates the conditional mean of a dependent variable given a set of explanatory variables. Quantile regressions can be used to characterise the entire conditional distribution of the dependent variable. The regression coefficient associated with an explanatory variable is interpreted as the marginal change in the given conditional quantile of the dependent variable corresponding to the marginal change in the variable. Comparisons of coefficients across different percentile levels allowed us to infer the effects of a certain variable at different points in the SF–6D distribution. Moreover, quantile regression is more robust to outliers. ^{Reference Koenker19} We based the inference on the median (percentile 50). Mental disorders and chronic physical conditions were introduced together in the model, adjusting by those sociodemographic variables that were statistically significant (P<0.20) in the bivariate Kruskal–Wallis tests. Additionally, we tested all the first-degree interactions. We compared a model with several first-degree interactions to the model with no interactions and it did not significantly improve the model fit so, following the parsimony principle, we decided not to include them. Third, to conduct sensitivity analyses, we performed eight additional quantile regressions with the inference based on the first (10), second (20), third (30), fourth (40), sixth (60), seventh (70), eighth (80) and ninth (90) deciles. We were also able to construct trend charts with the coefficients. That is, we were able to examine how the entire distribution changed after taking participants' characteristics into account. Interquartile range (IQR) regressions between extreme deciles (10 and 90) and between first (25) and third (75) quartiles were carried out with the aim of assessing whether the adjusted coefficients of the mental disorders and chronic physical conditions were different. We conducted bootstrap (b = 500) estimates of standard errors of regression coefficients for each decile regression.

Finally, the HRQoL loss associated with each of the disorders was estimated by multiplying the marginal effect of the conditions that reached significant criteria by the prevalence of the condition. This can be interpreted as the magnitude of the burden measured in annual loss in QALYs without considering mortality. ^{Reference Saarni, Suvisaari, Sintonen, Pirkola, Koskinen and Aromaa2} All analyses were carried out with the STATA 10 software for Windows. All significance tests were performed using two-sided tests evaluated at the 0.05 level of significance.

SF–6D index by sociodemographic characteristics

Table 1 shows the median on the SF–6D score by sociodemographic characteristics. Women had a lower SF–6D score than men (0.82 v. 0.72, P<0.0001). The median SF–6D score decreased progressively with age, except in the oldest group. Regarding marital status, those previously married had the lowest SF–6D score. Those in paid employment but on sick leave had a median of 0.61; lower than those in paid employment (0.82). The SF–6D score was 0.73 for those participants in the lowest educational group and 0.80 for those in the highest.

SF–6D score by conditions

The median SF-6D score varied by disease group, from 0.55 for participants with any mood disorders to 0.74 for participants with diabetes or high blood pressure (Table 2). When considering specific illnesses, major depressive disorder showed the lowest SF–6D (0.53), followed by social phobia (0.60), dysthymic disorder (0.60), panic disorder (0.60) and migraines (0.66). Those with heart attack showed the highest SF–6D (0.76). The only conditions that were not statistically significant when presence and absence were compared were any substance misuse, alcohol dependence/misuse, heart attack and diabetes. The worst health state profile (i.e. 34555) was the most frequent in participants with any mood disorder.

Impact of mental disorders and chronic physical conditions on the SF–6D

Figure 1 shows the modifying effect of mental disorders and chronic physical conditions when adjusting for sociodemographic characteristics and morbidity. The x-axis shows the decile on which the inference was based. Mood disorders, anxiety disorders and chronic pain were always statistically significant, regardless of the decile used. It was observed that any mood disorder was the condition category that had the greatest impact on the SF–6D score, independently of the decile used in the quantile regression. The trend chart demonstrated that the predictive capacity of mood disorders increases along with SF–6D score. That means that the differences in the HRQoL between participants with or without mood disorders is more evident among those with higher SF–6D scores. The IQR regression between extreme deciles (ninth and first) confirmed this. The coefficient of this regression (difference between ninth and first coefficients) was –0.083 (95% CI –0.115 to –0.052, P<0.0001). This statistical difference was also found between the first (25) and third (75) quartiles (coefficient –0.072, 95% CI –0.094 to –0.049, P<0.0001). Chronic pain also had an impact on the SF–6D score but conversely: as the SF–6D score increased, its predictive capability decreased. That is, differences in the extreme IQR regression (between first and ninth decile) showed a difference between coefficients of 0.052 (95% CI 0.015–0.061, P<0.0001). This difference was also found for chronic pain in the IQR regression conducted with first and third quartiles (0.040, 95% CI 0.034–0.071, P = 0.002).

Fig. 1 Impact of mental disorders and chronic physical conditions in the Short Form–6D index, adjusted for sociodemographic characteristics, by percentile (non-parametric interquartile range).

The impact of the other conditions was independent of the decile chosen. In fact, the extreme IQR regressions (between first and ninth decile) did not show any statistical differences for anxiety disorders, cardiovascular diseases, high blood pressure, diabetes or 12-month any substance misuse. When considering the first and third quartiles for the IQR regression, high blood pressure showed a statistically significant trend comparable to that of pain: as SF–6D increased, its predictive capability decreased (difference 0.021, 95% CI 0.001–0.040, P = 0.04). The pseudo R ₂ for each quantile regression oscillated between 0.1266 for the quantile regression with the inference based on the ninth decile and 0.2168 for the quantile regression with the inference based on the median. Table 3 shows the complete model with inference based on the median.

Burden of disease in primary care

Table 4 shows the annual QALY losses per 100 000 primary care patients that could be explained by each condition. Chronic pain was associated with the greatest QALYs loss, followed by mood disorders (5064 and 2634 respectively). Diabetes showed the smallest QALYs loss (250). Quality-adjusted life-year losses for substance use disorders were not calculated as they did not reach statistical significance in the quantile regression with the inference based on the median.

Strengths and limitations

One of the strengths of this study is that it has been conducted using a large representative sample of primary care attendees with good external validity. Moreover, to the best of our knowledge, there are few other studies with sufficiently large epidemiological samples in primary care to compare the impact of both chronic physical conditions and mental disorders in annual QALY losses. Another important strength is the statistical strategy used that permitted exploration of the predictive power of the different mental and physical conditions according to different SF–6D values. Additionally, the 36-item Short Form Health Survey (SF–36) ^{Reference Ware and Sherbourne20} and the different versions of it have been recently recommended by STAKES as a routine patient-centred outcome measure ^{Reference Korkeila Jyrki21} and this paper shows the different uses of this kind of measure. Finally, this study was carried out in a Mediterranean country. Previous studies aiming to study the burden of disease have been conducted in Anglo-Saxon or Scandinavian countries and little is known about the burden of diseases in southern European countries. It is well known that the expression of mental disorders differs across cultures, ^{Reference Breslau, Javaras, Blacker, Murphy and Normand22} so our paper also represents a chance to study how robust previous results on burden are.

Some limitations should be mentioned. First, we used UK tariffs and some cultural bias may be expected. A re-analysis of these data should be carried out when Spanish tariffs are available. Second, we estimated QALYs without adjusting for the years lived with the conditions. Therefore our estimates should be considered with caution as there could be a bias. Third, we did not consider the ‘treatment effect’. The vast majority of participants were receiving treatment that may have modified their previous health status. The utility assigned to their health status could be overestimated because they are receiving some form of care. Fourth, chronic physical conditions were ascertained using a checklist rather than an examination by a physician. However, it should be borne in mind when considering this limitation that checklists have been found to provide useful information about both treated and currently untreated chronic conditions, ^{Reference Knight, Stewart-Brown and Fletcher23} and they can predict out-patient healthcare use, hospitalisations and mortality. ^{Reference Fan, Au, Heagerthy, Deyo, McDonell and Fihn24} Additionally, methods research has shown that self-reporting of chronic physical conditions shows moderate to high agreement with medical records data. ²⁵ Fifth, our checklist did not include some serious illnesses such as cancer or neurological disorders that also have a big impact on HRQoL. Although some information about these illnesses was gathered in the open question included at the end of the checklist, the low prevalence of both illnesses and the different methodology used to assess them meant that they were not included in our analyses. We did not consider, as a part of the burden, the family burden and/or the social burden (stigma) that are associated with both physical and mental disorders. The burden associated with mental disorders could be underestimated as it is well documented that these conditions increase the burden on informal caregivers. ^{Reference Jungbauer, Bischkopf and Angermeyer26} On the other hand, we have to take into account that mental disorders were assessed considering the previous 12-month period, whereas chronic physical conditions were assessed over the lifetime period. Thus, it is possible that the impact of chronic physical conditions was underestimated, as these conditions are treated more by physicians.

Comparison with previous findings

Our data are similar to those obtained in other studies. However, statistical analyses usually consider parametric approaches making comparisons difficult, as our method has different assumptions. Moreover, most of the studies reviewed considered utilities derived from the EQ–5D. Different studies have shown that utilities obtained could vary according to the instrument used (EQ–5D or SF–6D). Lamers et al point out that ‘both discriminated between severity subgroups and captured improvements in health over time. However, the use of EQ–5D resulted in larger health gains and consequent lower cost-utility ratios, especially for the subgroup with the highest severity problems’. ^{Reference Lamers, Bouwmans, van Straten, Donker and Hakkaart27}

Consistent with our results, in a study carried out in the general population aiming to measure HRQoL decrement and loss of QALYs associated with pure and comorbid forms of depressive and anxiety disorders and alcohol dependence, Saarni et al ^{Reference Saarni, Suvisaari, Sintonen, Pirkola, Koskinen and Aromaa2} found that mood disorders had the worst impact on HRQoL. This was also found in a survey carried out in the general population in Sweden, ^{Reference Burstrom, Johannesson and Diderichsen28} where depression showed the lowest utility (0.38). One study aiming to assess changes in depression utilities before and after treatment showed that baseline depression utility was 0.33. ^{Reference Sapin, Fantino, Nowicki and Kind29} After treatment, this utility increased to 0.85 among treatment responders, to 0.72 among those who partially responded and to 0.58 among those who did not respond to treatment. This last value is similar to the one we obtained (0.53), where participants with depression could be in ongoing treatment for their condition or receiving treatment for other conditions if their depression was not detected or treated. Values from Germany for depression utility ranged between 0.56 and 0.64. ^{Reference Günther, Roick, Angermeyer and König5} Lastly, Revicky & Wood reported utilities for depression (stratifying by severity) derived from the SF–36. ^{Reference Revicki and Wood30} Utilities varied from 0.55 to 0.63 for moderate depression, from 0.64 to 0.73 for mild depression and from 0.72 to 0.83 for antidepressant maintenance therapy.

It is important to note that the impact of mood disorders in the SF–6D increases as the index does, suggesting that mood disorders are conditions sensitive to small losses in quality of life. This could be explained by the fact that the dimensions comprising the index are very close to the symptoms of depression: role limitation, mental health (feeling downhearted), social limitations and loss of vitality.

With regard to chronic pain we found an overall median utility of 0.72, higher than those previously reported (range 0.52–0.61). ^{Reference Ruchlin and Insinga31} Assessment methods or the different sample used could explain these differences. On the other hand, Brown et al, who studied the utility associated with migraine, found a utility of 0.62 (assessed with the Health Utilities Index 3 (HUI3)). ^{Reference Brown, Neumann, Papadopoulos, Ruoff, Diamond and Menzin32} This value is close to the 0.66 we found.

Similar to our results, a study focusing on respiratory conditions reported utilities ranging from 0.63 (in a sample of individuals with non-controlled asthma) to 0.80 (in individuals whose asthma was controlled). ^{Reference Szende, Svensson, Stahl, Meszaros and Berta33} However, this study used the SF–6D derived from the SF–36 and was carried out in a sample in ongoing rehabilitation.

The utility for high blood pressure was 0.74. This value was very similar to that found by a study conducted in the Swedish general population, which reported utilities for high blood pressure ranging from 0.73 to 0.81. ^{Reference Bardage, Isacson, Ring and Bingefors34}

Regarding diabetes, a Canadian study demonstrated that its utilities oscillated between 0.88 (when it presents without comorbidity) and 0.77 (when it is comorbid with other illness). ^{Reference Maddigan, Feeny and Johnson35} We found similar results: diabetes showed a utility of 0.74.

Contrary to the WHO report, ^{Reference Mathers and Loncar9} heart attack did not have a high impact in QALY losses. This could be related to the fact that we did not take into account quantity of years lived with the condition, nor mortality, which could increase the number of QALY losses.

Estimation of QALY losses showed that mood disorders ranked second, behind chronic pain. This may be explained by the high prevalence of the latter condition in our sample. However, we have to take into account that mood disorders showed a five times smaller prevalence than that of pain disorders and that QALYs loss for mood disorders were only half that of pain, highlighting the importance of mood disorders in disease burden. For example, if we compare QALY losses of mood disorders with those of cardiovascular disease, which showed similar prevalence in our sample, QALY losses associated with mood disorders were nearly ten times higher that those associated with chronic cardiovascular conditions. This elevated quantity of QALY losses associated with mood disorders could also be explained by the fact that mood disorders affect all dimensions forming the health profile, and that the worst profile (345555) is more frequent among people with depression.

In conclusion, our findings show that mood disorders are responsible for a large percentage of QALYs lost in Catalan primary care patients, slightly below that of chronic pain. As our data on mental disorders prevalence is very similar to previous reports, ^{Reference Üstun and Sartorius36} we think that our results can be generalised to other populations. The considerable expense that depression generates at both the individual and societal level justifies investment in strategies designed to reduce these costs. General practitioners are in a privileged position to detect and treat depression, and every effort should be made to improve training for these professionals.