Study population
The study population is based on the Finnish Twin Cohort, which consists of twin pairs born before 1958, with both twins alive in 1975, at the time of the first cohort-wide health questionnaire [13]. A second cohort-wide health questionnaire was sent in 1981, and a follow-up questionnaire was mailed in 1990 to the 16,179 twins born in 1930–1957 who had replied to either of the previous questionnaires. Of these twins, 12,502 subjects (77.3 %) responded [14]. For the first two questionnaires, response rates were 89 and 84 %, respectively.
The study is limited to questionnaire and anonymised medical record data, for which permission from the relevant authorities has been sufficient. The use of the questionnaire data for record linkage studies has been approved by the ethics committee of the Department of Public Health, University of Helsinki. The twin cohort members were informed about the use of the questionnaire information to study genetic and environmental influences on common diseases and their predictors and also about the record linkage, and been informed that they may withdraw from the study at any time.
The screening method for an epidemiological estimation of subjects with potential fibromyalgia
In our earlier study [12], based on the 1990 survey data, we classified these subjects into three latent symptom classes based on their answers to certain questions aiming to detect fibromyalgia. The latent class (LC) analysis is a statistical system which classifies data in categories that are internally as homogeneous and externally as heterogeneous as possible in relation to specific qualities, in this case to these FM-questions. From the original survey questions based on the Yunus et al. criteria suggestions from 1989 [15], we chose these questions about FM symptoms and FM-associated symptoms (from now on all called FM symptoms) according to the American College of Rheumatology 1990 classification criteria [1]. The questions were: How often have you had these symptoms during the last year? 1) stiffness in the limbs and trunk in the morning, 2) stiffness in the limbs and trunk in the evening, 3) pain and stiffness in the neck, 4) soreness with touching of the neck, back, trunk, or limbs, 5) numbness in the extremities, 6) daytime tiredness. 7) Does the stiffness in the limbs and trunk in the morning last “less than 15 min”, “about half an hour”, or “about 1–2 h”? and 8) Does low air pressure (rain, snow, storm) worsen the pain in the trunk or limbs?
Questions 1 to 6 were provided with the alternatives “never”, “daily or nearly daily”, “3–5 times per week”, “1–2 times per week”, “about once a month”, and “more seldom”. Exclusion of subjects with missing data concerning these classification questions yielded 10,608 subjects. Three latent classes represented the subjects best: LC1 indicated no or few symptoms; LC2 had some symptoms; and LC3 had a high frequency of FM symptoms resembling clinical FM patients. In LC3, over 80 % of the subjects had morning stiffness, 58 % had tender points, 65 % had neck pain and stiffness, and 48 % had daytime tiredness at least three times a week.
We also used the answers of a group of 49 clinically-diagnosed FM patients [12] as a validation data set. The LC method described classified all clinical FM patients to LC3 [12].
Exclusions
In the present study, we aimed to identify predictors of onset of LC2 and, in particular, LC3 memberships, i.e. predictors of the high incidence of FM symptoms as a proxy for WSP (widely used as a screening method) and FM. As these symptoms were not assessed in the 1975 and 1981 questionnaires, we used available information on pain-associated conditions to exclude those who would very likely be classified into LC2 or 3 as shown in the flowchart (Fig. 1). Thus, we first excluded the subjects having inflammatory rheumatic diseases or malignancies (which might cause the symptom load) by linking the data with national Special Refund Category data in the Drug Reimbursement Register held by the Social Insurance Institution and with data on incident cancers from the Finnish Cancer Registry to 1993. Subjects with missing data concerning particular regional pain questions either in 1975 or 1981 were excluded. We then excluded the subjects with possible FM symptoms at baseline by excluding those who reported pain in the neck and shoulders and back at the same time, in either 1975 or 1981 (this served as a proxy for pain at multiple sites, referred to as “WSP” in Fig. 1). Earlier, we had shown that regular use of analgesics was almost entirely restricted to the LC3 subjects. Therefore we also excluded those who reported having used analgesics of that frequency (on at least 180 days per year) in either 1975 or 1981 (Fig. 1). The final baseline sample for analysis, i.e. the population at risk, we presumed to be free of pain at multiple sites and free of major conditions giving rise to similar symptoms.
Study variables
We selected the variables analysed as potential predictors based on reports in the literature on putative predictors; that is, variables that had positive associations with FM or WSP in either cross-sectional or prospective, population-based, qualified studies mentioned in this article [2–11]. These included separate regional pain symptoms, headache, migraine, sleeping, underweight, overweight or obesity as indexed by BMI, smoking, and physical activity.
These potential predictors were assessed as follows. A simple question “In the last years, have you had pain in the back, shoulders, or neck that has impaired your working capacity?” with the alternatives “yes” and “no” for each region, assessed regional pain symptoms. The occurrence of migraine was based on the subjects’ report of diagnosed migraine [16]. The occurrence and frequency of headache was assessed in 1981 only by the question “Do you have headache?” with the options “daily or almost daily”, “many times a week”, “about once in a week”, “about once a month”, “many times in a year (but not every month)”, “once in a year or less” and “practically never” (reference category) [16]. Another question was “Do you usually sleep well?” with the options “well”, “fairly well”, “fairly badly”, “badly”, and “cannot answer” [17]. As the number of individuals reporting sleeping “badly” at baseline was small (36 and 52 in 1975 and 1981), we merged the two first alternatives into the category “good sleep” (reference category) and the next two into “poor sleep”. The number of “cannot answer” replies was also small (44 and 48 in 1975 and 1981) and they were handled as missing data.
The self-reported height and weight in both 1975 and 1981 produced BMI values which were classified in four categories based on WHO criteria: at least 30 (obese), at least 25 but less than 30 (overweight), under 18.5 (underweight), and at least 18.5 but under 25 (normal weight, reference category).
Two questions were on leisure-time physical activity. The question about year-round leisure-time activity, with the alternatives 1) “I do not exercise in my leisure time practically at all”, 2) “a bit”, 3) “fairly”, 4) “fairly much” and 5) “much”, originally assessed physical activity. As we wanted to look at physical passivity as a possible predictor and activity as a potential protecting factor, we re-classified these replies into three categories: physically passive (1–2), physically active (4–5), and average (3), which was used as a reference. Leisure-time exercise frequency was originally assessed by the alternatives 1) less than once, 2) 1–2 times, 3) 3–5 times, 4) 6–10 times, 5) 11–19 times, and 6) more than 20 times a month [18]. For this study we re-categorised these as three alternatives: 1) at most twice, 2) 3–10 times (reference category), and 3) at least 11 times a month. For smoking status, the subjects were classified into four groups: current smoker, former smoker, occasional smoker, and non-smoker (reference category) [19].
Reported gender, age (calculated from registry data and the date of response to the query), and education presented potential confounders. Education was originally reported with nine alternatives from elementary school to college or university degree. For this study, we calculated the mean of the education years for each alternative to form a continuous variable.
Zygosity was diagnosed by a validated questionnaire method using questions on similarity in appearance and confusion by strangers [20].
Statistical analysis
Two sets of analyses were conducted, first among all individuals (as a standard cohort analysis) and secondly within twin pairs to adjust for unmeasured genes and other factors shared by siblings.
Cohort analysis of individuals
For the analysis of potential predictors for FM symptoms among individuals, we used multinomial logistic regression analysis with the three latent symptom classes as the categories of the dependent variable. The asymptomatic class, LC 1, served as the reference category. As potential predictors, we analysed back pain, neck pain, shoulder pain, headache (data available only from 1981), migraine (data available only from 1981), sleeping problems, physical passivity or activity, BMI, and smoking, using the information from 1975 to 1981. The bivariate associations were first tested with logistic regression analysis, adjusting for age and gender. Based on these analyses, (i.e. including those variables with significant associations), we performed multivariate logistic regression analyses, adjusted for gender, age, and education. These variables were taken forward to the within family, i.e. pairwise analyses described below.
Before the pairwise analyses, we used the standard cohort approach to test for any possible moderation effect. We thus included interaction terms for gender, age, and education in the analysis with the data stratified by gender (men vs. women), age (under vs. over the median age), and education (high-school education, yes vs. no).
Post-hoc analysis of individuals
To analyse whether the persistence or recurrence of regional pain (back, shoulder, and neck pain) had any effect on the association with the future symptom class, we made re-analyses in a sub-sample with those individuals who had replied to the questionnaires both in 1975 and 1981, comparing positive reports at both time points (persistent or recurrent pain) to positive–negative combinations (pain at only one time point) and negative reports at both time points (no pain) in all three regions.
We did an additional analysis considering only those with more recent pain onset by inclusion of those subjects reporting no pain (back, shoulder, or neck) in 1975.
Within-pair analysis
The second set of analyses used the information on twinship to assess the possible effect of genetic or familial environmental factors on the relationship between predictors and FM symptoms. We identified the twin pairs discordant for “latent class” status, i.e. pairs in which one twin was classified as LC3 and the co-twin as LC1. If the genetic or environmental factors shared by the twin pair could account for the relationship between predictors and FM symptoms, the risk (the ORs) would presumably decrease. If particularly the genetic factors could account for the relationship, some association would appear in dizygotic (DZ) twins (who share on average 50 % of their segregating genes) but not in monozygotic (MZ) twins (who have an identical genotype) [21]. For this assessment, we used conditional logistic regression analysis. The predictors that were significant in the multivariate model were all included in this analysis.
Statistical software
We used the Stata version 12 in the pairwise analyses; in all other analyses we used SPSS version 19.
