Sample
We used data from the third Chilean National Health Survey carried out between August 2016 and March 2017 (ChNHS 2016-2017). It corresponds to a nationally representative sample of people older than 15 years, structured as a multistage complex design. Interviews and collection of laboratory and anthropometric measures were accomplished by trained individuals in at least two visits to the household of participants. Several quality control check points were implemented. The overall response rate was 67.0%, which corresponds to 6,233 respondents. The survey was commanded by the Chilean Ministry of Health and performed by the Department of Public Health of the Pontificia Universidad Católica de Chile with ethical approval from the Ethic Committee of the same university.
Musculoskeletal disorders and pain
We included six major musculoskeletal conditions associated with chronic pain, all capable to be identified through the ChNHS 2016-2017, namely: chronic low back pain, chronic shoulder pain, hip and knee osteoarthritis, fibromyalgia and chronic musculoskeletal pain. The latest is a broad diagnosis which encompasses multiple MSKD and was recently added to the 11th version of the International Classification of Diseases [11]. The main questionnaire used in this survey to collect information about musculoskeletal disorders was the Community Oriented Programme for the Control of Rheumatic Disease Core Questionnaire (COPCORD-CQ) [12]. This instrument collects information about the presence of ‘pain, stiffness, sensitivity or bone, muscle or joints swelling’ in 22 body regions. All musculoskeletal disorders considered only cases with pain during the last 7 days lasting at least 3 months in the body region of the disorder. Since the respondents could point out different pain locations, for chronic low back pain, chronic shoulder pain, and hip and knee osteoarthritis, we choose only cases with a declared preferential location of pain in the body region of the disorder. For chronic musculoskeletal pain we selected cases with pain in any of the 22 locations explored by the COPCORD-CQ, restricting the cases to those with intensity of pain ≥ 3/10 using a visual analogue scale. For fibromyalgia, hip and knee osteoarthritis, cases associated to a traumatic cause of pain were excluded. For fibromyalgia we attempted to meet the American College of Rheumatology 2010 criteria [13]. However, since these criteria include symptoms that were not explored by COPCORD-CQ, we were forced to use other questionnaires available in the ChNHS 2016-2017 to fulfil them. Cognitive symptoms, unrefreshed sleep and somatic symptoms were extracted from items of a disability questionnaire, which asks the following questions: ’Due to your health, how difficult was to remember things or concentrate?’, ‘Due to your health, how difficult is it to sleep?’ and ‘Due to your health, how much difficulty did you have in feeling any physical pain, such as back pain, stomach pain, or headache?’, respectively [14]. Fatigue symptoms were extracted from an item available in the CIDI-SF questionnaire (see below), which formulates the following questions: ‘During those same two weeks [of depressive symptoms], did you become more tired or with less energy than usual?’. Fibromyalgia was defined as ≥ 7 pain locations and a score ≥ 5 for other symptoms, or between 3 and 6 pain locations and a score ≥ 9 for other symptoms, restricting cases to those who had pain intensity ≥ 3/10.
Other variables used in the analysis
For description and adjustment purposes, we included other variables extracted from the ChNHS 2016-2017 in the analysis. They include; age; sex; marital status (married/cohabiting, annulled/separated/divorced, widowed, single); education (more than 12 years, between 9 and 12 years, less than 9 years of formal schooling); occupation (working for salary, looking for work, working without salary, and Not working and not looking for); and three prevalent comorbidities, hypertension, diabetes and depression, all of them frequently associated to musculoskeletal disorders [15]. Hypertension was defined as blood pressure ≥ 140 mmHg systolic and/or ≥ 90 mmHg diastolic after five minutes of rest, or normal blood pressure but self-report of diagnosis and treatment (i.e., lifestyle modifications or under drug treatment). Similarly, diabetes was defined as fasten glycemia ≥125 mg/dl or normal glycemia but self-report of diagnosis and treatment (i.e., oral hypoglycaemic agents or insulin). The detection of cases with a depressive episode during the last 12 months the survey used the Composite International Diagnosis Interview short form (CIDI-SF) questionnaire [16, 17] applying the DSM-IV criteria [18].
Burden of Disease
We calculated the number of Disability Adjusted Life Years (DALY) [19, 20] due to the six selected musculoskeletal disorder for the year 2017, in Chile. This metric results from the sum of years of life lost by premature death (YLL) and years lived with disability (YLD) due to a certain disease. For the set of selected musculoskeletal disorders, we assumed zero YLL, and any record of them was attributed to error of misclassification. The number of YLD were calculated through the multiplication of the number of prevalent cases with a particular sequel of the disease and a disability weight for that specific sequel, and then adding up the marginal results of the different sequels of the disease. Disability weights were extracted from the Global Burden of Disease study [21], and its values can range between 0 (absence of disability) and 1 (death). The description of sequels by each MSKD and the disability weights are available in the supplementary material. Smoothed prevalence of each sequel according to single ages for each sex were obtained using a backward selection method on a regression model that initially included a quadratic and cubic functional from for age and an interaction term between age and sex. Predicted values for single ages were calculated and then multiplied by the intercensal population expected for Chile in the year 2017 [22], through a Monte Carlo simulation (10,000 replications), assuming beta distribution for the prevalence. The number of cases for each sequel and each single age were collapsed in 18 strata of age and sex, assuming gamma distribution. Then, the prevalence for each stratum was recalculated dividing them by the intercensal population. The adjustment by sequels was carried out using the COMO procedure described elsewhere [23]. Briefly, for each stratum we simulated a population of 5,000 individuals with the probability of presenting each sequel equal to their prevalence, assuming independency between sequels. A total disability weight was calculated for each simulant and used to adjust the disability associated to each sequel. The procedure is repeated 1,000 times in order to propagate the second order uncertainty around parameters, assuming beta and binomial distributions as appropriate. The result of the COMO procedure is a YLD rate, which is multiplied by the intercensal population of the strata, to obtain the YLD due to the sequel. The YLD from different sequels are added to obtain the YLD due to a musculoskeletal disorder, assuming a gamma distribution. People younger than 15 years was assumed with 0 DALYs due to musculoskeletal disorders.
Loss in health state utilities
The health state utilities (HSU) were calculated using the EQ5D questionnaire, which inquiries about the ‘statements [24] best describe your own health state today’ through 5 items with three Likert alternatives. It explores the following domains of functioning: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression [25]. The combination of answers can describe 243 different health sates, which are transferred to a continuous scale anchored between values equal to 0 (equivalent to death) and 1 (perfect health) using social preferences previously estimated for Chilean population through a time trade off protocol [26]. Because we are interested in exploring the loss of HSU (also referred as disutilities) we transformed the HSU according to: LHSU = (1 – HSU), where 0 means perfect health, and 1 means death. On some occasions, duly indicated, for better interpretation, we preferred to use a scale between 0 and 100 (e.g., reporting coefficients from regression models). The LHSU attributable to each musculoskeletal disorder, at individual level, was estimated using an ordinary least square regression model where the dependent variable was the LHSU and the independent variable was the condition, adjusted or not by sociodemographic variables, other musculoskeletal disorders, and comorbidities (i.e. hypertension, diabetes and depression). Using the same regression model, we predicted values of LHSU to each individual separately by people with the musculoskeletal disorder (LHSU1) and without (LHSU0). Also, we predicted values of LHSU for people with the musculoskeletal disorder assuming a counterfactual scenario where they do not have the condition, i.e., imputing zero in the dummy variable that marks the presence of the musculoskeletal disorder (LHSU1’). The total LHSU attributable to each musculoskeletal disorder, at population level, was calculated as the sum across n individuals (i) of the subtraction between LHSU1i and LHSU1 i’ (i.e., [LHSU1i – LHSU1’i]). Finally, the fraction of the total LHSU in the population produced during 2017, attributable to each musculoskeletal disorder was calculated as: ( [LHSU1i – LHSU1i’])/ ( [LHSU1i] + [LHSU0i]), which is equivalent to the epidemiological concept of population attributable fraction. Similar procedure but for dichotomous outcomes has been described previously [5]. The extended methodology has been recently submitted by Zitko P. et al.
Pain domain of loss of health utilities
Since the EQ5D questionnaire explores the health state using different domains of functioning, and one of them is ‘pain and discomfort’, it was possible to calculate the counterfactual scenario where no one reports pain and discomfort. This was accomplished imputing the lowest value in the variable of the pain and discomfort item of the EQ5D, and then recalculating the LHSU. In that way it is possible to calculate the fraction of the LHSU that are attributable to the domain of pain and discomfort: (observed LHSU – LHSU assuming no pain of discomfort) / observed LHSU. The calculus can be performed in population with different musculoskeletal conditions, and for all domains of the EQ5D.
All statistical analysis was performed using the software R 3.5.0, and its package survey. The main functions used to calculate the burden are available in the supplementary material.