|Home | About | Journals | Submit | Contact Us | Français|
This paper describes methods used in the project “Orofacial Pain Prospective Evaluation and Risk Assessment” (OPPERA) and evaluates socio-demographic characteristics associated with temporomandibular disorders (TMD) in the OPPERA case-control study. Representativeness was investigated by comparing socio-demographic profiles of OPPERA participants with population census profiles of counties near study sites and by comparing age- and gender-associations with TMD in OPPERA and the 2007-09 US National Health Interview Survey. Volunteers aged 18-44 years were recruited at four US study sites: 3,263 people without TMD were enrolled into the prospective cohort study; 1,633 of them were selected as controls for the baseline case-control study. Cases were 185 volunteers with examiner-classified TMD. Distributions of some demographic characteristics among OPPERA participants differed from census profiles, although there was less difference in socio-economic profiles. Odds of TMD was associated with greater age in this 18-44 year range; females had three times the odds of TMD as males; and relative to non-Hispanic-Whites, other racial groups had one-fifth the odds of TMD. Age- and gender-associations with chronic TMD were strikingly similar to associations observed in the US population. Assessments of representativeness in this demographically diverse group of community volunteers suggest that OPPERA case-control findings have good internal validity.
The project “Orofacial Pain: Prospective Evaluation and Risk Assessment” (OPPERA) was designed to identify a broad range of genetic, physiologic, psychosocial and clinical characteristics that influence development of painful temporomandibular disorders (TMD). Implicitly, putative risk factors should be understood within the context of socio-demographic factors associated with TMD. Hence, OPPERA’s first specific aim was to determine whether socio-demographic characteristics generally considered predictive of TMD are associated with elevated risk of first-onset TMD and with increased odds of chronic TMD.
Because they are fundamentally related to causes of ill-health, age, gender, race, and ethnicity feature prominently in the design and analysis of virtually all epidemiologic studies. Some studies limit the potential impact of demographic characteristics by restricting enrollment to one demographic subgroup or by matching when selecting cases and controls. Other studies apply statistical adjustments for demographic characteristics during data analysis. The scientific rationale stems from the fact that demographic characteristics determine exposure to a range of risk factors and protective health factors, and therefore might confound associations between those factors and the condition under study. Demographic characteristics also are at the core of public health policy. Age-related patterns of ill-health are critical when projecting health care needs in aging populations. The US Healthy People initiative lists reductions in racial and ethnic group health disparities as a primary objective for the nation.30
Socio-economic status has a profound impact on many diseases and disorders, including an array of pain conditions such as musculoskeletal pain, sciatica, ulcer, and neuropathic pain.23 Socio-economic health inequalities have been characterized as “large, persistent, and even increasing” in developed countries.27 An important finding is that socio-economic inequalities are not confined to people who live in poverty.21 Instead, they manifest as a gradient in disease risk that increases among progressively lower rungs of the socioeconomic ladder.
Given that demographic and socio-economic influences on health are pervasive and the use of socio-demographic variables is orthodox, it is informative when exceptions to predominant trends are found or when there is uncertainty about the very nature of the association. Past studies of TMD provide examples of both. In the 2002 US National Health Interview Survey (NHIS), prevalence of TMD-like pain was greater in non-Hispanic-Whites than other racial-ethnic groups,11 making it one of the few disorders that occurs less frequently in racial and ethnic minorities. In its request for applications that led to the OPPERA study, the National Institutes of Health stated: “The highest prevalence [of TMD] is observed in young adults (18-43 years)”.6 However, in one of the few examination surveys of a representative population sample, prevalence of TMD symptoms was greatest among 50-59 year olds.12
The specific aims of this paper are to:
The OPPERA project comprises: (i) a prospective cohort study of first-onset TMD, (ii) a baseline case-control study of chronic TMD, (iii) a matched case-control study of incident TMD, and (iv) a prospective case-cohort study of the course of TMD (Figure 1). Collectively, these observational studies were designed to identify risk factors for onset and persistence of TMD. Methods for all four studies are described below, although results from OPPERA in this paper and others in this issue come solely from the OPPERA baseline case-control study mentioned above.
The OPPERA project was reviewed and approved by institutional review boards at each of four study sites and at the data coordinating center, Battelle Memorial Institute. All study participants verbally agreed to a screening interview by telephone and provided written consent for all other study procedures. The OPPERA study is being conducted under the auspices of a Certificate of Confidentiality (NIDCR-06-17) between the National Institutes of Health and Dr William Maixner, Principal Investigator of OPPERA. The Certificate protects the privacy of research subjects.
Study participants were recruited from communities in and around four US academic health centers: (1) The University of Maryland at Baltimore, MD, with a nearby population 651,154; (2) The University of Buffalo, NY, with a nearby population of 292,648; (3) The University of North Carolina at Chapel Hill, NC, with a nearby population 48,715; and (4) The University of Florida at Gainesville, FL, with a nearby population of 95,447. (All population figures are from the 2000 US decennial census.) At baseline, all participants attended their local study sites for data collection by questionnaire, clinical examination, measurement of pain perception and autonomic function, and collection of a blood sample. Follow-up data collection will continue through 2012. All data and biological samples are managed by the study’s data coordinating center, Battelle Memorial Institute, through its offices in Durham, NC.
The prospective study of first-onset TMD recruited a cohort of adults aged 18-44 years who did not have TMD when examined at their baseline visit. Study participants continue to be followed to identify onset of TMD. Follow-up questionnaires, completed once every three months, identify people who develop significant symptoms suggestive of TMD, with examinations then performed to determine presence or absence of TMD in those reporting significant symptoms.
Recruitment took place between May 2006 and November 2008, when healthy volunteers were sought from communities in and around each study site. We sought to enroll individuals who varied in their risk of developing TMD and who covered the spectrum of major socio-demographic groups seen in counties near the study sites. However, there was no purposeful random sampling of residents in those counties.
The goal was to recruit 800 people without TMD from each study site. Volunteers were recruited using advertisements in local newspapers and radio stations, emails distributed through university networks, flyers posted in and around academic health centers, and word of mouth. Recruitment materials explained that healthy volunteers aged 18-44 years were needed for a study of muscle and jaw function. Respondents telephoned the local study site, at which point research personnel explained the study, administered a computer-assisted telephone interview to screen for study eligibility, and scheduled a clinic visit for people who met selection criteria.
Prior to their baseline clinic visit, potential enrollees completed 16 in-home questionnaires (Supplementary e-Table 1), either online or on paper forms that were mailed to them. The questionnaires asked about past experiences of potential risk factors and trait-like psycho-social characteristics thought to predict risk of TMD. Accompanying manuscripts describe specific details about the origins, reliability, and validity of these and other data collection procedures.
At the start of the 3-hour baseline clinic visit, study procedures were explained in detail and study participants signed the consent form approved by the study site’s IRB. They completed seven more questionnaires evaluating state-like psychological characteristics and symptoms with timeframes anchored to the clinic visit (Supplementary e-Table 1).
A trained and calibrated examiner then assessed the head and neck to confirm absence of TMD and to measure clinical characteristics thought to predict risk of TMD. The examination protocol, which was based on the Research Diagnostic Criteria for TMD,9 is described in detail in an accompanying paper [Ohrbach et al., in this volume]. To summarize, examiners first identified the relevant orofacial region by touching the following anatomical areas bilaterally: temporalis, preauricular, masseter, posterior mandibular, and submandibular areas. Examiners then asked a series of structured questions to evaluate symptoms regarding pain history, focusing on pain location, its history, and whether or not it was of orofacial origin. Signs of TMD were then assessed by determining responses to jaw movement and digital palpation of orofacial structures. The origin of any pain reported during those procedures was classified by the examiner into one or more of 10 anatomical locations, each considered bilaterally: temporalis, masseter, lateral pterygoid, submandibular, and temporomandibular joint (TMJ).
Quantitative sensory testing was performed after the examination to measure responses to standardized noxious stimuli delivered using pressure pain, mechanical cutaneous (pricking) pain, and heat pain. Full details of those measurements are provided in Greenspan et al. [included in this volume]. Autonomic function was monitored at rest, during orthostatic challenge, and during the Stroop color-word test and pain-affect test. Full details of those measurements are provided in Maixner et al.[Autonomic paper, in this volume.] Anthropometric measurements were recorded, and a 20ml sample of peripheral blood was collected by venipuncture for subsequent DNA purification and genotyping.
At the end of the clinic visit, instructions were provided for a set of 28 daily pain diaries to be completed during the subsequent four-week period. Study participants were told they would be contacted for quarterly health updates described in more detail below. They received $100 payment for completing all baseline data collection procedures ($20 for pre-clinic questionnaires, $50 for physical assessments at the clinic, $20 for providing a blood sample, and $10 for the 4-week pain diary).
The clinical examiners from each study site were trained and calibrated together at University of Buffalo; calibration was made to a single expert dentist [YG] based at the Buffalo site who served as the reference examiner throughout the reported study period. Expert dentists from each site were calibrated according to the case classification decision rules. Three reliability studies were conducted in which examiners from each site conducted pairs of blinded, replicated examinations of non-OPPERA volunteers: one examination in each pair was conducted by the OPPERA examiner, and another by the reference examiner. The first reliability study regarding examiner performance was conducted at Chapel Hill; the study included 24 subjects (15 cases, 9 controls), and each of the clinical examiners and reference examiner independently performed the examination on each subject. At that study, the clinical examiners reviewed findings with their respective expert dentist, just as they would during actual study conduct, and the two of them determined case classification. Over the next year, the examiner and expert dentist from each site, in turn, traveled to Buffalo in order for the expert dentists to be calibrated by the reference examiner. Immediately after calibration, the study site’s pair of clinical examiner and expert dentist was evaluated for reliability, as judged against the reference examiner. At approximately 12-15 month intervals, repeat reliability studies were conducted in Buffalo. The overall Kappa reliability coefficients for the reliability in TMD case classification were: 0.87 (2006), 1.0 (2007), and 0.96 (2009).
Study-wide inclusion and exclusion criteria were evaluated in the telephone interview (Supplementary e-Table 2). In summary, eligibility criteria were aged 18-44 years, fluent in English, not receiving orthodontic treatment, and not pregnant or nursing, and had negative responses to each of 10 questions about significant medical conditions and no history of facial injury or surgery,.
To be eligible for the prospective cohort study, potential participants had to report absence of significant orofacial pain symptoms and they had to be free of TMD when examined (Supplementary e-Table 2). Symptoms were identified first in the telephone interview where all four of the following criteria had to be met:
The symptom criteria regarding orofacial pain were subsequently verified during structured questioning at the beginning of the clinical examination. After defining the relevant orofacial region by touch, examiners verified that participants had experienced pain of orofacial origin in the defined orofacial region for no more than 4 days in the prior 30 days. Furthermore, controls had to be classified by examiners as having neither myalgia nor arthralgia. Consistent with RDC criteria, signs of pain in response to jaw movement and digital palpation were not sufficient to classify TMD and, therefore, people who reported pain during examination procedures remained eligible for enrollment in the prospective cohort study.
The reference period of 30 days used in the examination was shorter than the three-month reference period of the telephone interview due to the increased complexity of defining the relevant anatomy and identifying the orofacial origin of pain in the examination inquiry. The intention was to restrict examinations to people whose initial telephone response signified a relatively long period with few or no symptoms. The clinic visit could then provide an opportunity for more in depth probing of the most recent pain to permit valid reporting of potentially overlapping pain, such as headache.
Follow-up is continuing through mid 2012, with the expectation that approximately 300 people will be identified with first-onset TMD. Follow-up is by quarterly health update (QHU) questionnaires, administered once every three months, either online or on paper. Outside of that quarterly schedule, QHU questionnaires can also be administered to participants who contact their local study site describing symptoms. People who respond positively to QHU questions about TMD symptoms are asked to repeat most of the same questionnaires used at baseline and they are invited to the study site’s clinic for clinical examination, quantitative sensory testing, measurement of autonomic function, and collection of a blood sample. A positive QHU response is based on criteria that are similar to the symptom questions used to recruit TMD cases for the case-control study (described below). To be classified as an incident case, these study participants must then be classified with TMD myalgia, arthralgia or both, using the same RDC-based examination criteria described below for TMD cases in the case-control study. Further details about identification of onset-cases will be reported in later papers that investigate incidence of TMD.
The unmatched case-control study aimed to recruit 200 people with examiner-classified TMD myalgia, arthralgia or both (“TMD cases”). Controls were a 50% random sample of people without TMD who were enrolled in the prospective cohort study. The case-control study used data collected at baseline assessments for these TMD cases and controls.
Recruitment of TMD cases reported in these papers occurred from May 2006 through November 2008. An additional round of recruitment began in April 2010, with the aim to enroll an additional 900 chronic cases of TMD for the purpose of a genome-wide association study. Enrolment will continue through July 2012, and data from those TMD-cases will be reported in later papers. Recruitment of cases used the same procedures described above for the prospective cohort study. The goal was to enroll people representing the major socio-demographic groups found in counties nearby the study site. Specifically, enrollment of TMD cases was not limited to people who were seeking care for TMD.
Data collection procedures for TMD cases were identical to those described for the prospective cohort study.
To be enrolled as TMD-cases, volunteers had to meet the same study-wide inclusion criteria used in the prospective cohort study (Supplementary e-Table 2). However, symptom and examination criteria differed with the aim of selecting people who had experienced facial pain symptoms for at least six months and who had examiner-confirmed classification of TMD myalgia, arthralgia, or both.
To conform with the symptom criteria, potential TMD cases first had to report pain with sufficient frequency in the telephone interview; that is pain in the cheeks, jaw muscles, temples or jaw joints during the preceding six months (at least 15 days in the preceding month and at least five days per month in each of the five months preceding that [Supplementary e-Table 2]). The symptom criteria were then verified during a clinical examination, following the same protocol described for the prospective cohort study. For potential TMD cases, examiners confirmed that orofacial pain experienced during the preceding month was in the relevant orofacial area. If pain was described as headache located in the temporalis region, it was considered a sufficient symptom if it occurred in the relevant orofacial area during at least five days. However, pain in the orofacial area that was attributable to other causes, for example toothache, was not a sufficient symptom. The reference period of 30 days used in the examination was shorter than the three-month reference period of the telephone interview for the same reasons described for selection of participants in the prospective cohort study.
In addition to symptoms of TMD, reports of pain in response to examination procedures were required to classify people as TMD cases. Specifically, TMD cases had to report pain in response to movement or palpation in at least one of the two TMJs or at least three of the eight orofacial muscle groups (each assessed bilaterally): temporalis, masseter, lateral pterygoid, and submandibular.
A subset of people without TMD was selected from among all enrollees in the prospective cohort study using a stratified random sampling procedure. The rationale was to create a “reserve” sample of controls whose data could be combined with TMD cases who have been recruited since April 2009 for additional case-control analyses. Potentially, the reserve sample could also be used for validation of exploratory data analysis procedures conducted among the primary group of controls—for example, to validate findings from cluster analysis. Using the SAS SURVEYSELECT procedure, the cohort of 3,263 people who did not have TMD was stratified according to study site, and a simple random sample of 50% of enrollees was selected from within each site.
The number of people enrolled as controls was selected to meet statistical power requirements for the prospective cohort study. The baseline case-control study recruited 185 people with chronic TMD, comparing them with 1,633 controls drawn from the enrollees in the prospective cohort study. The smaller number of cases was the primary determinant of statistical power for hypotheses addressed in the case-control study. For binary predictor variables, odds ratios as small as 1.7 could be detected with 80% power, assuming type I error of P=0.05, and exposure prevalence as low as 15%. For continuous predictor, the minimum-detectable difference in standardized means between 185 cases and 1,633 controls was 0.22 standard deviations, corresponding to a minimum-detectable standardized odds ratio of 1.25, again assuming type I error of P=0.05 and power of 80%.
OPPERA investigators convened five data analysis working groups to develop data quality guidelines and to guide statistical analysis for five types of data: clinical assessments; psychological characteristics; quantitative sensory testing; autonomic function; and genetics. For continuous measures, quality assessment began with generation of box plots, separately for cases and controls, to identify potential outliers. Global editing rules were created to exclude values that were beyond the range of measurement or that were outside physiologic limits. Individual data forms with anomalous values were inspected visually to identify possible recording or scanning errors, and corrected when feasible. Accompanying papers describe specific quality control steps used for analysis of each set of putative risk factors.
In addition to examiner training and calibration procedures described above, three data quality assessment steps were used to assure quality of TMD case-classification:
For putative risk factors measured as continuous variables, summary measures were computed from responses to multiple items that formed subscales. Summary measures were calculated first for people with “complete data,” meaning those for whom valid values were recorded for each item comprising a subscale or other summary measure. Typically, those measures yielded continuous scores that were analyzed to generate means, standard deviations and quantiles describing their distribution.
For putative risk factors measured as continuous variables, means were compared between TMD cases and controls and analysis of variance was used to test for statistical significance of the difference in means, adjusting for study site. Analysis of variance was used even for variables with markedly non-normal distributions because, in large samples, those least-squares methods are valid for any distribution.19 Analysis of variance was used as well in instances where the variances of the two groups were unequal since the procedure is robust even when this assumption is violated. Exceptions were made when the precedent was to transform the variable to improve its interpretability; for example, some measures of heart rate variability were computed on the log scale.
For characteristics that were measured categorically, mutually-exclusive groups were created and proportions were compared between cases and controls. This approach was used in addressing the fifth aim of this paper. In some instances, ordinal variables were managed in the same way, although some ordinal variables with many levels were treated as continuous variables (described above). The statistical significance of any differences in proportional distribution between cases and controls was evaluated by creating dummy variables for all but one reference category and evaluating the set of dummy variables using the likelihood ratio statistic from a binary logistic regression model, where the dependent variable was TMD case status. The model additionally adjusted for study site as a covariate. Parameter estimates and standard errors from the model were used to generate odds ratios and 95%CIs for the association between the characteristic and odds of TMD. As done for the continuous measures, a second logistic regression model generated odds ratios and 95%CIs that additionally adjusted for three demographic characteristics: age in years (modeled as a continuous variable), a single dummy variable for gender, and a single dummy variable signifying race/ethnicity as non-Hispanic-White or other.
All P-values were computed without adjustment for multiple tests, and P=0.05 was therefore not nominated as a threshold for statistical significance. In this paper’s case-control analyses, 11 socio-demographic characteristics were investigated so Bonferroni correction for the probability of type I error would yield a critical P-value of 0.05 ÷ 11 = 0.004. Using the same rationale, rejection of the null hypothesis concerning odds ratios would occur only if the 99.6% confidence interval excluded the null value of one. In general, though, drawing conclusions about statistical significance of associations should be avoided, even with correction for multiple tests, because these papers report only univariate- or demographically-adjusted results. Instead, judgments about statistical significance will be made in subsequent papers using multivariable modeling to consider multiple characteristics simultaneously, as proposed in the OPPERA heuristic model (see Maixner et al., Overview paper, in this volume).
To address aims 3 of this paper, selected socio-demographic data items from the 2000 US decennial census were tabulated and the percentage distributions were compared with percentages and corresponding 95% confidence intervals among controls in the OPPERA baseline case-control study. The ratio of the OPPERA percentage relative to the Census percentage was used as an indicator of the degree to which people in the OPPERA prospective cohort study were representative of the major socio-demographic groups within the same age range in counties nearby each study site. Ratios of >1.5 and <0.7 were used as thresholds to signify marked departure from representativeness. Relevant counties nearby recruitment sites are shown in Supplementary e-Figure 1. They were:
County-specific data were selected from files downloaded from the US Census Bureau:28
For income, the census data were reported only for two relevant age-groups (25-34 and 35-44 years), whereas other variables were reported for three relevant age groups (18-24, 25-34, and 35-44 years).
OPPERA data were from the 1,633 study participants in the prospective cohort study selected at random for the baseline case-control study. They reported relevant socio-demographic characteristics in the telephone interview and the self-completed demographic questionnaire. People with invalid or missing responses to OPPERA questions were omitted from analysis for that item. Within each study site, percentage distributions were tabulated permitting comparison between census benchmarks and participants in the OPPERA prospective cohort.
To address the sixth aim of this paper, the age- and gender-patterns of association with TMD in the OPPERA baseline case-control study were compared to age- and gender-related patterns of association with self-reported TMD-like symptoms in the US population. US population estimates were generated using data from the 2007-2009 NHIS, a nationally representative cross-sectional survey that interviews, annually, a sample of the civilian, non-institutionalized US population. Datasets for each of the three years were downloaded from the Centers for Disease Control and Prevention.29 This analysis was limited to people aged 18 years or more. Cases of TMD-like symptoms were defined as individuals who answered “Yes” to the following question: “During the past three months, did you have facial ache or pain in the jaw muscles or the joint in front of the ear?” The 69 individuals who responded “don’t know” or who did not answer the question were excluded from the analysis, resulting in a sample size of 72,836 people. Age was defined as a categorical variable with seven levels: 18-24, 25-34, 35-44, 45-54, 55-64, 65-74, and 75+. The prevalence of TMD-like symptoms, cross-classified by age group and gender, was computed using sampling weights that were part of NHIS’s complex multistage survey design. Standard errors were estimated using the Taylor-series approximation method.18 Since three years of NHIS data were pooled, weights from the pooled data were divided by three to avoid overestimation. All calculations were performed using the SAS SURVEYFREQ procedure.
Telephone interviews were conducted with 5,781 people who expressed interest in the project identifying 4,057 who were eligible for the prospective cohort study. Of those, 3,350 (83%) attended the clinical assessment (Supplementary e-Table 3). All but 87 were confirmed as eligible during the clinical assessments, yielding 3,263 people without TMD who were enrolled into the prospective cohort study. Random sampling selected 1,633 of those people whose baseline data were used for the baseline case-control study. The telephone interview identified an additional 280 people who were putative cases of TMD, of whom 202 (72%) attended the clinical assessment. Examiners confirmed TMD in 185 of them, and they constituted the sample of TMD cases enrolled into the baseline case-control study.
For both cases and controls, word of mouth was the most commonly-reported way in which enrollees heard about the OPPERA study, followed by email, posted flyers and advertisements (Table 1). Only 7% of TMD cases and 9% of controls said that they had been referred from a research clinic.
The number of enrolled TMD cases varied more than two-fold among study sites, while there was less site-variation, in relative terms, in enrollment of controls (Table 2). Baltimore enrolled more people aged 35-44 years than either of the other two age groups, whereas 35-44 year olds represented the smallest age group at the other sites. Baltimore had nearly equal numbers of males and females, while females were a clear majority at each of the other three study sites. Baltimore had the largest number of Black/African American study participants, Gainesville had the largest number of Hispanics, and Buffalo had the largest number of people in other racial/ethnic groups, many of them Asian.
Demographic characteristics are cross-classified in Table 2 to provide additional details about study participants at each study site.
At the Baltimore study site, age distributions were generally similar for controls in the baseline case-control study and people living in nearby counties (Table 3). In contrast, at the other study sites, OPPERA participants had a younger age distribution than people living in nearby counties. Percentages of males and females did not differ markedly between OPPERA study sites and the nearby counties. In Baltimore, there was a markedly higher proportion of Blacks/African Americans in OPPERA than in nearby counties (percentage ratio = 2.4) — a discrepancy that was not observed at other study sites. Hispanics of both gender were over-represented at the Florida site, and male Hispanics were over-represented at Buffalo. For most of the OPPERA percentage estimates, 95% confidence intervals were approximately ±5%, which did not markedly alter these interpretations regarding representativeness.
At all study sites, OPPERA participants were more likely to have never married than people living in nearby counties. At most study sites, OPPERA participants were more likely than residents of nearby counties to have educational qualifications beyond high school, although at Baltimore, the OPPERA sample was fairly representative with respect to educational attainment. In Baltimore and Buffalo, OPPERA participants had a lower income distribution than households in nearby counties. At other study sites, were generally small differences in most categories of income, although all three sites were over-represented with people in the highest income category.
Among people classified with TMD in the baseline case-control study, more than 90% reported facial pain that began at least one year before enrollment into the study, with the majority reporting a history of at least three years (Table 4). Pain was experienced in recurrent bouts by most. Nearly three-quarters of TMD cases reported episodes of pain that lasted for at least 15 days in the preceding month, and approximately one-half had individual episodes that usually lasted at least one day. Most had seen a health care provider for their pain at some time, although only about one-third had done so within the preceding six months.
Using the Characteristic Pain Intensity scale from the Graded Chronic Pain Scale, cases registered a mean pain intensity of moderate (51/100) as integrated over the present and prior six months (Table 5). Cases also registered mean levels of approximately 2/10 with respect to interference in each of daily activities, social activities, and work activities due to facial pain. The mean number of days cases were prevented from attending to their usual activities was 11, though the distribution was skewed, with the maximum reported value at the scale maximum of 180. Cases registered mean pain intensity of approximately 7 and unpleasantness of nearly 8 using the Gracely scale for pain intensity and unpleasantness. Responses on both scales were skewed toward the maximum score of 20.
Distributions of all three demographic characteristics differed markedly between TMD cases and controls (Table 6). Odds of TMD increased across successively older age groups, and females had more than three times the odds of TMD as males. Relative to non-Hispanic-Whites, other racial groups had lower odds of TMD, with fully-adjusted odds ratios ranging from 0.2 (for Black/African American and for Other groups) to 0.4 (for Hispanics). Because of the overall similarity of odds ratios for specific minority groups, their data were pooled, yielding a fully-adjusted odds ratio of 0.2 (95%CI = 0.2, 0.3) for the combined group of non-Whites. Odds ratios for individual demographic characteristics tended to be consistent, whether or not they were adjusted for other demographic characteristics.
While country of birth was not associated significantly with occurrence of TMD, people whose first language was not English had half the odds of TMD relative to people whose first language was English (Table 7). However, the latter effect ceased to be statistically significant after adjustment for age, gender and race. Higher educational attainment was associated with greater odds of TMD, and although odds ratios were attenuated after adjustment for demographic characteristic, there remained an approximate two-fold increase in odds associated with any level of education beyond high school (Table 7). In contrast, there was a conspicuous lack of association between TMD and income, satisfaction with socio-economic position, and health insurance coverage.
Population estimates from the NHIS for females revealed an inverted-U relationship between age and prevalence of TMD-like symptoms (Figure 2A). Prevalence reached 7.1% among women age 35-44 years whereas it was 5.1% in women aged 18-24 years and 3.5% in women aged 75 years or more. Prevalence in men was about half that of women in the corresponding age group. The age-gradient among men followed a weak inverse relationship, although age variation in men was less apparent than in women.
Very similar patterns of age- and gender-related variation were observed when the proportion of TMD cases was plotted for the OPPERA baseline case-control study (Figure 2B), albeit within the truncated age distribution of OPPERA study participants. Specifically, among females, the proportion of cases increased in successively older age groups, while among males, there was little age-variation. As observed with the NHIS data, the gender difference in proportion of TMD cases was greatest among OPPERA participants aged 35-44 years, and was much smaller among 18-24 year olds.
This paper addresses OPPERA’s first scientific aim regarding socio-demographic associations with chronic TMD, and it provides information to gauge representativeness of the study groups. Representativeness is important for two reasons. First, case-control studies should select controls who represent the same population from which cases arise.25(p 116) This is one critical requirement for internal validity of associations estimated in case-control studies. Case-control studies are internally valid when “comparison groups are selected and compared in such a manner that the observed differences between them may, apart from sampling error, be attributed only to the hypothesized effect under investigation”, while sampling error is “that part of the total estimation error of a parameter caused by the random nature of the sample”.16 For example, when cases are recruited from clinics, it is essential that controls be selected in such a way that, if they developed the condition, their potential to be detected as cases at those clinics would be similar to that of cases. The OPPERA project sought volunteer participants from communities in and around academic health centers. Fewer than 10% of cases reported hearing about OPPERA through research clinics; the majority of cases heard either by word of mouth or from advertisements, flyers and emails. This result, coupled with findings regarding clinical pain reports, symptom profiles and history of treatment among TMD cases, is consistent with TMD as it occurs in the community at large, rather than in treatment-seeking TMD cases. Likewise, controls reported hearing about OPPERA primarily by word of mouth or from advertisements, flyers and emails. Age-and gender- associations with TMD observed in the OPPERA baseline case-control study were very similar to the associations observed in the US population, suggesting that the methods used to select cases and controls did not seriously distort estimates of demographic associations with chronic TMD. Taken together, these suggest that both cases and controls were selected from a community-based population of volunteers, and that the methods of selection produced good internal validity.
Representativeness of the OPPERA controls is of additional importance to gauge generalizability of findings that will emerge from the prospective cohort. Generalizability, also termed external validity, occurs when unbiased inferences regarding a target population can be inferred from associations observed among participants in a specific study16. A target population is a larger group of people than those studied, and usually refers to people living in similar places (e.g., an entire county, state or nation) and at similar times as the study sample. In OPPERA, controls were a non-probability sample of volunteers, rather than people who agreed to participate after being selected at random from a sampling frame. Statistical sampling theory dictates that measures of association and other statistics computed from non-probability samples cannot be construed as population estimates.1(page 116) Even when random samples are selected, non-participation and loss-to-follow-up is inevitable, and it is important to evaluate representativeness of the study participants. A common technique is to compare characteristics of study participants with population benchmarks. Despite not having used probability sampling methods, we also investigated representatives by comparing socio-demographic characteristics of OPPERA controls with Census 2000 data. Not surprisingly, there was evidence of non-representativeness, most notably for age distribution, marital status, and educational attainment. Furthermore, the degree of non-representativeness varied among study sites. In the case-control study, different numbers of cases were enrolled at each site for operational reasons. We therefore adjusted for study site and demographic characteristics to produce internally valid estimates of association in the case-control study. However, it is important to note that those statistical adjustments do not guarantee external validity.
Generalizability is not essential when investigating putative risk factors in epidemiological studies. Even in studies that lack generalizability, a characteristic found to be risk factor provides “proof of principle” that the characteristic can influence health. Furthermore, if a study enrolls sufficient numbers of people in major demographic groups, the “proof of principle” potentially can be evaluated separately in those groups. Indeed, it is unusual for prospective cohort studies to use probability sampling methods, and doing so might even threaten internal validity, for example by making it more difficult to obtain uniformly accurate measurement among all study participants.25(p 146)
Demographic diversity was an explicit recruitment goal for OPPERA, in part because the project was funded by a request for applications that required inclusion of “subjects from both genders, all racial and ethnic groups (and subgroups)”. When the study was designed, relatively little was known about age-related and racial/ethnic variations in rates of examiner-determined TMD for US community-based samples drawn from non-clinical settings. For that reason, it was important that an unmatched case-control study design be used, and that cases and controls reflected demographic variation in the United States. Hence, the OPPERA sites were selected because they included large metropolitan areas and smaller rural communities.
Another goal of recruitment for the prospective cohort study was to select people who reported no history of TMD but who varied in characteristics that likely were associated with risk of developing the condition. Specifically, it was important not to exclude “sub-clinical” individuals who had some (but not sufficient) signs and symptoms of TMD. We therefore permitted enrollment of people who reported infrequent pain in the orofacial area (i.e., less than five days per month) or who reported some headaches (fewer than five episodes in the past month), provided they were not found to have TMD when examined. People were not enrolled in the prospective cohort study if they reported five or more headache episodes per month because that number of temple-region headaches potentially could be regarded as a variant of TMD.3 A variety of other overlapping pain conditions was permitted without qualification.
People with these sub-clinical and overlapping conditions were not excluded from the prospective cohort study for two reasons: (i) the aims of the prospective cohort study might be compromised by enrolling a highly pain resistant group which may have a very low incidence rate of first-onset TMD—the endpoint for the main scientific goals of OPPERA; (ii) sub-clinical and overlapping conditions are highly prevalent in the US population, which may have limited OPPERA’s generalizability by excluding people with those conditions. Likewise, for the case-control analysis, it is important to retain people with sub-clinical conditions, because their exclusion produces estimates of association that are biased away from the null.24
The OPPERA baseline case-control findings confirmed that TMD occurs more frequently in females than in males.8, 20 Although there is less published evidence about racial and ethnic group variation in TMD, these OPPERA findings are consistent with results from the NHIS showing that non-Hispanic whites have higher prevalence of TMD-like pain than racial and ethnic minorities in the US.11 Our own analysis of NHIS data revealed patterns of age- and gender-specific prevalence in the US population that were strikingly similar to patterns observed in the observed proportion of cases among the 18-44 year-olds in OPPERA. Together with results from an examination survey of women in New York,12 these findings demonstrate a positive association between age and occurrence of TMD among women within the 18-44 year age range. The findings clarify ambiguity in the literature regarding the association between age and TMD in early- and mid-adulthood. For example, a systematic review noted that most cross-sectional studies of TMD prevalence found “a negative linear relationship between prevalence and age”,20 a premise echoed by the National Institutes of Health in its request for applications6 that led to the OPPERA study.
Because of its rigorous sampling design and its large sample size, the NHIS represents the gold-standard for prevalence of self-reported health conditions in the United States. Prevalence of TMD-like symptoms in the NHIS was based on a single-item question that is similar to the question used in other national interview surveys.14 In the study of women in New York City,12 a comparable single-item screening question yielded prevalence (10.1%) that was very similar to prevalence of examiner-determined TMD (10.5%). This was despite the fact that the screening question had low sensitivity (42.7%), a limitation that was balanced by excellent specificity (94.7%). If the NHIS screening question for TMD had similar validity, the NHIS prevalence figures represent only slight underestimates of the true population prevalence. Taken together, the age- and gender-related associations with TMD in the OPPERA case-control study suggest that OPPERA should have broad applicability for the populations at these four US study sites.
Past studies provide inconsistent evidence regarding the relationship between socio-economic status and TMD.20 A questionnaire-based survey of 50- and 60-year-olds in Sweden reported significantly greater odds of orofacial pain for people in blue collar occupations than in white collar occupations, although there was no significant association with education.13 In the United Kingdom, self-reported prevalence of orofacial pain was elevated 1.5-fold for people living in areas of the country that were the most socio-economically deprived relative to people living in affluent areas.2 In that study, orofacial pain included reported pain in the jaw joints and other parts of the face, pain during jaw function, tenderness of facial muscles, and other forms of oral pain such as burning and shooting pain. These case-control findings from OPPERA showed a general lack of association between examiner-classified TMD and indicators such as income and health insurance, and only a modest increase in odds of TMD associated with greater educational attainment. Satisfaction with financial and material circumstanace reflect financial strain15 which is one aspect of subjective social status, a construct found to be informative in social epidemiologic studies of health inequalities. 26 Despite the fairly consistent null associations with chronic TMD observed using a number of socio-economic indicators, the relationship warrants more detailed investigation. For example, these analyses adjust only for demographic characteristics, yet other risk factor for TMD might have confounded socioeconomic associations towards the null. Also, results from the prospective cohort study should provide additional insight into effects of socioeconomic status on risk of developing TMD.
Nonetheless, the very absence of a marked inverse social gradient in TMD makes it one of the few health conditions to defy that trend. In contrast, reviews of other pain conditions report a consistent, inverse gradient of reducing prevalence in higher socio-economic groups.23 In the case of widespread bodily pain, a large, population-based prospective cohort study was able to attribute the inverse gradient to an array of psychological characteristics: psychological distress, anxiety, depression, illness behavior, somatic symptoms, sleep problems, and threatening life events.4
In summary, these finding from the OPPERA baseline case-control study revealed an array of demographic associations: TMD was more likely in older than in younger age-groups, in females than in males, and in non-Hispanic-Whites than in other racial/ethnic groups. This latter result, and the general absence of a socio-economic gradient in TMD, are counter to the patterns of health disparities seen in many other diseases and disorders.
Demographic associations with TMD were consistent with population benchmarks and with other studies, suggesting broad applicability of these OPPERA findings. Greater occurrence of TMD in non-Hispanic-Whites than in other racial/ethnic groups and the lack of a socio-economic gradient contradicts the disparities seen in many other health conditions.
The authors would like to thank the OPPERA research staff for their invaluable contributions to this work. In addition, we express our gratitude to the participants who have devoted time and effort in support of this research. We are grateful to Andrew Munn of the Odum Institute at UNC Chapel Hill who created maps shown in Supplementary e-Figure 1.
This work was supported by NIH grant U01DE017018 and P01NS045685. This material was also supported with by the North Florida/South Georgia Veterans Health System, Gainesville, FL. The OPPERA program also acknowledges resources specifically provided for this project by the respective host universities: University at Buffalo, University of Florida, University of Maryland-Baltimore, and University of North Carolina-Chapel Hill. Roger Fillingim and Gary Slade are consultants and equity stock holders, and William Maixner and Dr. Luda Diatchenko are cofounders and equity stock holders in Algynomics, Inc., a company providing research services in personalized pain medication and diagnostics.
This work was done at: University of North Carolina at Chapel Hill, NC; University at Buffalo, NY; University of Maryland-Baltimore, MD; University of Florida, FL; and Battelle Memorial Institute, NC.
Portions of these data were presented at the 2010 Annual Scientific Meeting of the American Pain Society in Baltimore, MD.
Eric Bair, Kunthel By, Flora Mulkey, Cristina Baraian, Rebecca Rothwell, Maria Reynolds, Vanessa Miller, Yoly Gonzalez, Sharon Gordon, Margarete Ribeiro-Dasilva, Pei Feng Lim, Joel D Greenspan, Ron Dubner, Dawn Dampier, Charles Knott and Richard Ohrbach declare no competing interests.