Statistics from Altmetric.com
Gout is a chronic condition associated with impaired quality of life.1 The increased incidence of cardiovascular diseases in patients with gout has long been attributed to the strong association of gout and comorbid conditions such as diabetes, hypertension and dyslipidaemia.2 ,3 Large epidemiological studies have demonstrated that gout is a risk factor of cardiovascular diseases such as myocardial infarction, coronary heart disease and, above all, death.4 ,5
An accurate estimate of the prevalence of gout in the general population is needed to estimate the burden of the condition on the healthcare system. Moreover, at a national level, the assessment of the prevalence of gout would also allow for determining the healthcare needs of affected patients and thus facilitate the allocation of resources.6
Accumulating data support an increase in the prevalence of gout that may be attributable to recent shifts in diet and lifestyle, improved medical care and increased longevity.1 ,7 The most recent estimates of the prevalence of gout in Europe were 2.5% in the UK,8 1.4% in Germany,9 0.9% in Italy10 and as high as 3.9% in the USA.11 These prevalence estimates were based on self-reporting11 or on records from primary care databases8–10 rather than on microscopy identification of monosodium urate crystals, the gold standard for gout diagnosis.12 Both methodological approaches may have overestimated11 or underestimated10 the true frequency of gout in these countries. In addition, clinical diagnosis, used in other epidemiological surveys,1 ,13 has been shown to have poor sensitivity and specificity as compared with monosodium urate crystal identification.14 Because crystal diagnosis to confirm cases of gout is not practical at a population level, alternative survey methods are needed to estimate the prevalence of gout.
As compared with the prevalence of rheumatoid arthritis,15 spondyloarthropathies,16 and hip and knee osteoarthritis,17 that of gout in France is currently unknown. The prevalence of these diseases was successfully determined by a two-step telephone procedure with validated screening questionnaires.6 ,17 ,18
We aimed to determine the prevalence of gout in France via a telephone survey of a large representative sample of the general population (n=10 000). For this, we first needed to develop and validate a screening questionnaire suitable for use in the telephone survey. Here, we report the metrological performance of the developed screening questionnaire in a large case-control study including cases with crystal-proven gout.
This study was approved by the French Departmental Directorate of Health and Social Affairs, the Commission Nationale de l'Informatique et des Libertés, the French Data Protection Authority and the Comité Consultatif sur le Traitement de l'Information en Matière de Recherche dans le Domaine de la Santé. It was conducted in accordance with the Declaration of Helsinki and the Guidance for Good Clinical Practice. All participants gave their written informed consent to participate in the study.
Patients and methods
The case-control study was conducted in 14 rheumatology departments in France. The study enrolled patients >18 years old with a history of arthritis who had undergone arthrocentesis for synovial fluid analysis and search for crystals (monosodium urate and calcium pyrophosphate). Cases were patients with crystal-proven gout and controls were patients who had arthritis with effusion but without monosodium urate crystals in synovial fluid, and with no clinical diagnosis of gout.
Cases and controls were contacted by mail and asked to participate in the study. They were given an informed consent form explaining the aims and constraints of the study. If they did not respond to the mail, they were contacted by phone by a clinical research associate. A non-physician interviewer unaware of the identities of the respondents (case or controls) administered by phone the developed questionnaire to patients who agreed to participate in the survey. In addition, demographic characteristics were recorded in a case report form and data were entered in a database.
TB and MD designed a questionnaire suitable for use by patient interviewers. In total, 62 items (see online supplementary material) were organised in three sections: (1) demographic characteristics, comorbidities and treatments; (2) characterisation of the overall articular involvement (localisation, type); and (3) characterisation of the most prominent episode of arthritis (duration, intensity of pain, clinical features). Some items were obtained from published clinical criteria (Rome, New York, American Rheumatism Association (ARA), Mexico, the Netherlands).19 The questionnaire was designed to be easily administered over the telephone by an interviewer who was not a healthcare professional. This questionnaire took 15 min for administration. To evaluate its feasibility, we pretested its comprehensibility with a group of 20 patients.
The expected properties of the questionnaire were high sensitivity and specificity (number of true positives/number of cases and number of true negatives/number of controls). To estimate a 90% proportion with 5% precision, we needed 248 patients. For each item of the questionnaire, univariate comparisons between cases and controls involved Pearson's χ2 test (with Yates’ correction if required) or Fisher's exact test for categorical variables and Student t test or ANalysis Of VAriance (ANOVA) for continuous variables. Variables discriminating cases and controls at a type I error <0.10 were introduced in multivariate models. The probability of being a case was modelled by logistic regression with the best model chosen on the basis of the minimisation of the Akaike information criterion, measuring the loss of information when reduced in a model. Sensitivity, specificity, and positive and negative likelihood ratios were calculated, and the area under the receiver-operating characteristic curve was evaluated by the Harrel's c value, an approximation of the area under the receiver-operating characteristic curve for item performance.6 The Harrel's c value varies between 0 and 1; the closer to 1, the lower the performance error. The variables selected after this first step were introduced in a final model, with further selection with the Akaike information criterion. In addition, we used a classification and regression tree (CART) to find the best combination of variables able to classify cases and controls. CART is a stepwise process. At each step the population is divided into two branches and each resulting node contains a greater proportion of cases or controls. Nodes become more refined with progressive division. Pruning methods were used by cross-validation with the Gini criteria for allowing or not further division of a given node. Sensitivity, specificity, and positive and negative likelihood ratios were calculated. Statistical analyses involved use of SAS V.9.1.3 (SAS Institute, Cary, North Carolina, USA) and R statistical software (http://www.R-project.org, the R Foundation for Statistical Computing, Vienna, Austria) and the library rpart.
Cases and controls
Of the 520 patients (292 cases and 228 controls) identified within the 14 rheumatology departments and who were contacted by mail, 244 patients (102 (35%) cases and 142 (62%) controls) agreed to participate in the study and answer the questionnaire. The mean ages of cases and controls were 59.8±12.5 years and 61.4±14.2 years, respectively, (p=0.35). All cases had crystal-proven gout. Controls had rheumatoid arthritis (n=43), acute crystal pyrophosphate arthritis (n=30), osteoarthritis (n=22), peripheral spondyloarthritis (n=17) and other miscellaneous rheumatic disorders (septic arthritis, adult onset still disease, Lyme disease, undifferentiated arthritis; n=30).
Items related to comorbidities, self-reported diagnosis and treatments
Cases were more frequently men (p<0.0001) and had obesity (p<0.0001), hypertension (p=0.002) and cardiovascular disease (p=0.02) (table 1). In all, 92.9% and 85.6% of cases and 18.3% and 19.6% of controls self-reported a diagnosis of gout or hyperuricaemia, respectively (p<0.0001 for both). A total of 78.4% of cases and 10.6% of controls reported they had taken urate-lowering agents (p<0.0001), whereas 88.9% and 30.7%, respectively, reported the use of colchicine (p<0.0001); 62.9% of cases and 7.2% of controls reported tophus (p<0.0001).
Items related to the most prominent episode of arthritis
Cases and controls were asked to recall their most prominent episode of acute arthritis (table 2). Globally, it was more sudden (p=0.04), painful (p=0.002) and more often associated with inflammatory signs in periarticular soft tissues (p=0.001) in cases than in controls. In addition, it involved more often the first metatarsophalangeal joint and the feet in cases (p<0.0001 for both) but the hips or shoulders (p<0.001), hands (p=0.0018) or spine (p=0.0085) in controls. Consumption of alcohol (p=0.005) and heavy meals (p=0.002) were most often triggers in cases than controls. Treatment with colchicine was more frequent in cases than controls (p<0.0001), whereas controls more often used corticosteroids (p<0.0001), non-steroidal anti-inflammatory drugs (NSAIDs; p=0.015) or analgesics (p=0.011). Finally, resolution of pain occurred sooner in cases than controls (p<0.0001).
Selection of variables in logistic regression analysis
The first logistic regression model retained nine items as predictors of cases of gout (table 3). In the multiadjusted model, self-reported diagnosis of gout and self-reported diagnosis of hyperuricaemia had the highest ORs: 21.6 (7.2 to 76.9) and 5.5 (2.0 to 16.3), respectively. Male sex, tophus and hypertriglyceridaemia were associated with probability of being a case. Two features related to the most prominent arthritic episode—high level of pain (>9/10) and involvement of lower limbs for the affected joint (toes, feet or ankles)—were also strongly associated with probability of being a case. Conversely, treatments with corticosteroids or NSAIDs were negative predictors.
Because therapeutic modalities for treating gout flares may vary among countries, we built a second logistic regression model without the variables treatment with corticosteroids or NSAIDs. Six variables were retained (table 3). Among them, four from the first model were retained (self-reported history of gout and hyperuricaemia, pain intensity and involvement of toes, foot or ankles); the two other variables were history of cardiovascular disease and pain resolution of <15 days after arthritis onset. The equation for this model is as follows:
Selection of variables with CART
We built a CART using the 11 variables retained by the two logistic regression models (figure 1). Only three variables were retained: self-reported history of gout and tophus and involvement of toes, foot or ankles during the most prominent episode of arthritis.
Diagnostic performance of the three models (two logistic regression models and CART)
The three models had similar performances (table 4). The first logistic regression model that included treatment for acute arthritis (NSAIDs or corticosteroids) had 88.0% sensitivity and 93.0% specificity, for 12.5 and 0.13 positive and negative likelihood ratios, respectively. The second model had 87.5% sensitivity and 89.8% specificity, for 8.6 and 0.14 positive and negative likelihood ratios, respectively. Finally, the CART model had 81.3% sensitivity, 93.7% specificity, and 12.8 and 0.20 positive and negative likelihood ratios, respectively. About 90% of patients were correctly classified with the three models (figure 2). Of note, the item self-reported gout had 5.1 and 0.09 positive and negative likelihood ratios, respectively.
Given the preceding steps of selection, from the 62 items included in the initial questionnaire, we selected 11 items that could be used to distinguish gout from other rheumatological conditions (see online supplementary appendix 1). These 11 items take about 5 min to be administered.
In this study, we aimed to design and assess the performance of a telephone questionnaire for use by non-physician patient interviewers to estimate the prevalence of gout in France, which is currently unknown. With regression and CART analyses, we determined a set of 11 questionnaire items able to discriminate proven gout from other rheumatological disorders that could be used in a telephone survey of patients with a history of arthritis. The logistic regression models (sensitivity 88.0% and 87.5%; specificity 93.0% and 89.8%, respectively) and CART model (sensitivity 81.4%, specificity 93.7%) allowed for classifying 90.0%, 88.8% and 88.5%, respectively, of patients with gout.
The first logistic regression model retained nine items from an initial list of 62 administered questions, whereas the second model retained only six variables, including self-reported diagnosis of gout and hyperuricaemia combined with one comorbidity (ie, history of cardiovascular disease) and three typical features of acute gout (ie, high level of pain, involvement of toes, foot or ankle, and rapid resolution of symptoms). This set of six items has the advantage of not including any question related to treatment of flares. Indeed, management of flares may vary among countries20–23 and therefore, a questionnaire without items related to this issue might be more appropriate for comparing estimates of gout prevalence among countries.
As expected, self-reported diagnosis of gout was associated with the highest ORs in regression models: 21.6 (7.2 to 76.9) and 29.6 (10.9 to 93.1), respectively. Gout has often been defined by self-reporting in epidemiological studies, in particular in the National Health Interview Surveys.1 ,3 ,11 Although some authors found that self-reporting has acceptable reliability and sensitivity,24 others reported that it overestimates the prevalence of gout in surveys. Indeed, several studies have shown that the validation rate of self-reported gout ranges from 50% to 80% depending on the study design and type of classification criteria used.25–28 We found that self-reported gout had lower specificity (81.7%) and positive likelihood ratio (5.1) than with the two logistic regression and CART models, and the lowest rate (86.3%) for correct classification of gout. Therefore, the addition of a few items for self-reporting gout in a questionnaire could improve the specificity and performance to properly classify patients with gout.
Ideally, a screening questionnaire should be highly sensitive and specific, so that few patients with gout are missed and few patients without gout are identified as having gout. Our three models showed satisfactory sensitivity, specificity and positive likelihood ratio and compared well with other telephone questionnaires designed to estimate the prevalence of other musculoskeletal disorders.6 ,17 In particular, the positive likelihood ratios computed for the logistic regression and CART models were close to or >10, which shows their good diagnostic accuracy.29 Of note, we included in the control population about 20% of patients with acute calcium pyrophosphate crystal arthritis, formerly considered pseudogout,30 which mimics acute flares, to maximise false-positive errors that might also explain the relatively low specificity of self-reporting gout. This is likely to explain the high rate in controls of colchicine use for the treatment of the most prominent episode of arthritis, and the high prevalence of the self-reported gout frequency, as pseudogout is often misdiagnosed as acute attacks. Thus, the specificity of our models would be even higher in a general population in light of the low prevalence of acute calcium pyrophosphate arthritis.
As compared with the Janssens criteria31 or the clinical gout diagnosis criteria,32 our questionnaire was not designed for diagnostic purposes in primary care or secondary care settings. The Dutch criteria were developed as a diagnostic decision aid to help general practitioners with the diagnosis of gout in patients with monoarthritis and suspected gout.19 ,31 By contrast, our questionnaire aims to classify, with reasonable estimates, via a telephone survey, patients with gout or not in a large population sample. It was developed by interviewing patients and controls who had an episode of arthritis leading to joint fluid analysis, which allowed for the identification of patients with gout. The value of our questionnaire as a diagnostic tool for acute gouty arthritis is unknown.
In contrast to the Rome,19 the New York19and the ARA criteria,33 we used crystal-proven gout to define cases. Sensitivity and specificity of our questionnaire to detect gout seem higher than those of the ARA criteria which has been frequently used to select patients in clinical trials.34–36 Indeed, two studies have reported an external validation of the ARA criteria against a gold standard of synovial fluid (SF) analysis, as we did. In these works, the sensitivity was 70% and 80%, and the specificity was 79% and 64%, respectively.37 ,38
Our study has strengths and limitations. The characteristics of our patients might have affected the performance of the screening questionnaire. Patients with gout who agreed to participate were recruited from tertiary care hospitals and not a primary care setting, which may have led to selection bias, with an over-representation of severe gout, as illustrated by the high percentage of cases who declared having tophi. In addition, we cannot rule out that our patients with gout had a better knowledge of their condition than patients from primary care, and that this may have impacted the metrological performance of the questionnaire, However, conducting our case-control study in a primary care setting was unfeasible because synovial fluid analysis is seldom performed by general practitioners who treat most patients with gout.39 The wording of our questionnaire items might be a concern, because even minor changes in question content can influence prevalence estimates.40 Therefore, whenever feasible, we chose a wording previously used in studies involving interviews.3 For all these reasons, our questionnaire should be applied to an external set of patients for validation.
The strength of our work relies on the crystal-proven diagnosis of all patients with gout, monosodium urate crystal identification in synovial fluid being the gold standard for the diagnosis of gout.12 Finally, our questionnaire does not require current evidence of active joint inflammation, and thus can be administered in patients with intercritical or chronic gout. In conclusion, we developed a questionnaire suitable for use in telephone survey that demonstrates good properties for discriminating patients with and without gout. This questionnaire will be administered in a large sample of the general population to estimate the prevalence of gout in France.
The authors thank Philippe Ravaud for helpful discussion during the elaboration of the study protocol. The authors thank all the investigators and in particular Vincent Perez, Dominique Larzabal, Brigitte Palestro, Laure Perissin, Xavier Chevalier and all the clinical research associates who collected data for this study. The authors also thank the patients who agreed to answer the questionnaire.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.