Protocol and Registration

The protocol for this review was developed following the PRISMA-P reporting guideline ⁽²³ and was submitted for registration with the International Prospective Register of Systematic Reviews (PROSPERO, Centre for Reviews and Dissemination, University of York; and the National Institute for Health Research). To conduct this systematic review of diagnostic tests, the PRISMA-DTA checklist recommendations were followed ⁽²⁴.

Eligibility criteria

This review included diagnostic accuracy studies for the detection of TMD based on the FAI as an index test and the 3Q-TMD test as a comparator. For both tests, studies using the RDC/TMD and DC/TMD as the gold standard were examined. No restrictions were applied regarding the age or sex of participants or the publication language of the studies. Both “cohort” and “case-control” study designs were accepted.

As regards sample collection sites, studies conducted in both primary care (general dental centers) and secondary care (specialized orofacial pain centers) settings were included. To establish the target condition, patients in these studies underwent an individual clinical assessment of TMD (specifically joint noises, limitations in jaw movement, muscle pain, joint pain, and/or preauricular pain) following the criteria established by the INFORM consortium ²⁵.

Exclusion criteria were as follows:

Identification of study sources

The HIRU search filter for diagnostic accuracy studies was used ²⁶. An electronic search strategy was subsequently developed for PubMed and adapted for each of the following bibliographic databases: Web of Science, ScienceDirect, Scopus, and Scielo. Grey literature databases such as Google Scholar and ProQuest One Academic were also searched.

The search period covered the years from 1992 to April 2022. Additionally, the reference lists of included studies were manually reviewed to identify any additional relevant studies. A reference manager* was used to compile references and remove duplicates (*Mendeley®, Elsevier Amsterdam, The Netherlands).

Search strategy

Indexed terms and free terms were used to locate research conducted on TMD diagnostic studies, as well as their diagnostic accuracy. Below is a description of the search strategy using the filter for diagnostic accuracy studies ⁽²⁶:

Search: ((“temporomandibular joint disorders”(mh) OR “Disorder, Temporomandibular Joint”(tiab) OR “Disorders, Temporomandibular Joint”(tiab) OR “Joint Disorders, Temporomandibular"(tiab) OR “Temporomandibular Joint Disorder”(tiab) OR “TMJ Disorders”(tiab) OR “Disorder, TMJ”(tiab) OR “Disorders, TMJ”(tiab) OR “TMJ Disorder”(tiab) OR “Temporomandibular Disorders”(tiab) OR “Disorder, Temporomandibular”(tiab) OR “Disorders, Temporomandibular”(tiab) OR “Temporomandibular Disorder”(tiab) OR “Temporomandibular Joint Diseases”(tiab) OR “Disease, Temporomandibular Joint”(tiab) OR “Diseases, Temporomandibular Joint”(tiab) OR “Temporomandibular Joint Disease”(tiab) OR “TMJ Diseases”(tiab) OR “Disease, TMJ”(tiab) OR “Diseases, TMJ”(tiab) OR “TMJ Disease”(tiab)) AND (Diagnosis(mh) OR diagnosis(tiab) OR Triage(tiab))) AND (sensitiv*(TiAb) OR sensitivity and specificity(mh) OR (predictive(Tiab) AND value*(Tiab)) OR predictive value of tests(mh) OR accuracy*(Tiab)).

Study selection

In phase 1, two authors (HDR and CIR) independently evaluated the titles and abstracts of the identified studies by applying the previously established eligibility criteria. Once the articles were deemed eligible for inclusion, the reviewers performed a full-text screening in phase 2. Finally, articles that did not meet the eligibility criteria were removed. Although a third reviewer (MK) was available to intervene in the event of potential disagreements, these were settled among the authors in several consensus meetings.

Data collection process

Data collection was performed by the first reviewer (HDR) and verified by the second reviewer (CIR) to ensure the integrity of the contents. Disagreements were settled by consensus with the third reviewer (MK). A spreadsheet was used, and the following data were extracted for each included study:

Bibliometric data: year of publication, country of origin, author, publication journal.

Methodological data: total number of participants in each study, age, sex, patient blinding, and randomization method used. Study health care setting (primary or secondary care).

Statistical data: the following data were extracted: sample size of each group, sensitivity and specificity data, predictive values, diagnostic OR, and area under the ROC curve.

Risk of Bias

The risk of bias was independently assessed by two reviewers (HDR and CIR). The QUADAS 2 tool was used to evaluate the risk of bias and applicability of the studies ⁽²⁷. The evaluation covered four distinct domains: patient selection, index test, reference standard, and patient flow and timing throughout the study. Before applying the tool, a pilot test was conducted to ensure consensus on the assessment of risk of bias between the two reviewers. No overall summary score was calculated; however, for each domain, concerns regarding bias and applicability were rated as “low,” “high,” or “unclear.” Discrepancies were resolved through consensus. These results are fully detailed in an appended paper.

Diagnostic accuracy measures

Data from 2x2 tables were used to calculate sensitivity, specificity, and predictive values for each study. The results of individual studies were graphically presented by plotting sensitivity and specificity estimates (along with their 95% confidence intervals). Additional estimates, such as the diagnostic Odds Ratio, were calculated, and a proportional hazards model was used to estimate the performance of the tests on a receiving operating characteristic (ROC) curve. This approach provided a single value representing the overall detection ability of the screening tests.

Results Synthesis

The results were summarized by plotting estimates of sensitivity and specificity for the index and comparator tests (FAI and 3Q-TMD) on coupled forest plots and a matrix of receiver operating characteristics (ROC curves).

Meta-analysis

Summary operating points (summarized sensitivities and specificities) were estimated for both tests with a 95% confidence interval. Subsequently, the diagnostic odds ratio was determined to express the diagnostic accuracy of each test as a single number. Finally, the area under the curve for both tests was calculated on a receiver operating characteristic (ROC) curve using a proportional hazards model.

Study selection

A total of 798 articles were identified. Of these, 727 articles were excluded due to ineligible titles that did not align with the search objectives. At this stage, the main reason for excluding titles was that the studies did not pertain to screening tests. After selecting 71 studies for title and abstract evaluation, 49 duplicate studies were removed (phase 1). Ultimately, 22 studies were deemed eligible for full evaluation. After the full-text screening (phase 2), 12 studies were excluded because they did not focus on the evaluation of screening tests. Finally, 10 articles were included in the systematic review and meta-analysis. An overview of the selection process is presented in Figure 1.

Figure 1: 

Characteristics of the studies

The 10 included studies ^{(17, 19, 20, 28-34)} involved a total of 4,106 subjects (2,348 females, 580 males, 878 with unspecified gender), with a male-to-female ratio of 1 to 4. The age range of participants in these studies spanned from 11 to 78 years, with a mean age of 3.7 years.

The studies were conducted across 7 different countries, with sample sizes ranging from 102 to 923 participants. The descriptive characteristics of these studies are summarized in Table 1.

Table 1: Summary of the Main Characteristics of the Studies 

Risk of Bias and Applicability of the Studies

Due to the methodological differences among the included studies, it was necessary to assess the factors determining their internal and external validity. The QUADAS 2 tool ²⁷⁾ allowed to determine the risk of bias in the primary studies (internal validity). In this regard, none of the studies met all the methodological quality criteria. Five of them ^{(19, 31, 32, 34, 35)} were considered to have a high risk of bias, four were considered to have a low risk ^{(17, 20, 28, 36)}, and one was rated as unclear ⁽³⁵. The main issue was related to patient selection. Expert recommendations in diagnostic test accuracy (DTA) studies suggest that including patients selected as cases and controls should be avoided, as this can bias the diagnostic accuracy of a test. As shown in Table 1, only 4 studies included a consecutive sample of patients. Regarding applicability, or the extent to which the results can be generalized to other patients and settings, all studies were considered to have a low risk of bias.

Table 2 provides a summary of the different aspects analyzed in the risk of bias assessment.

Table 2: Risk of bias assessment 

References: ☺ Low Risk ☹ High Risk ? Unclear Risk

Results of Individual Studies

Despite the significant heterogeneity among the studies, both the 3Q-TMD test and the FAI had very good diagnostic accuracy. For TMD diagnosis, acceptable levels of sensitivity and specificity have been suggested to be at least 70% and 95%, respectively ¹⁴.

In contrast to diagnostic studies, screening studies focus on sensitivity, even at the expense of decreasing specificity, as the goal is to detect all suspicious individuals. The two studies on the 3Q-TMD test used the DC-TMD reference test and obtained the following diagnostic accuracy results: In Lovgren's study of the 3Q-TMD test in a primary setting, sensitivity was 0.81 (CI: 0.73-0.87), and specificity was 0.79 (CI: 0.73-0.83). In a secondary setting, sensitivity reached 0.96 (CI: 0.92-0.98), while specificity was 0.34 (CI: 0.28-0.40).

Regarding the studies that examined the FAI test, a high level of heterogeneity was observed. The results are categorized by the original version and the abbreviated version. Five studies on the original version ^{(19, 31, 32, 35, 36)} were identified, with most showing very good sensitivity and specificity data.

The studies on the short form of the FAI consist of three papers published between 2018 and 2021. The studies by Ujin-Yap and Pires ^17,34 demonstrated very good sensitivity and specificity (0.95-0.93 for the Yap study, and 0.86-0.95 for the Pires study). The study by Zagalaz-Anula ³⁷ showed acceptable sensitivity and specificity (0.78-0.79). Table 3 shows the results for sensitivity, specificity, and predictive values.

Table 3:  

References: 3Q-TMD Studies, FAI Studies, Abbreviated FAI Studies

Results Synthesis

Despite the limited number of articles, a high degree of heterogeneity was observed among the analyzed studies due to variability in sample characteristics, methodological differences, and risk of bias. The meta-analysis results showed excellent sensitivity values for the FAI and 3Q-TMD tests: 0.92 (CI: 0.88-0.95). Specificity values demonstrated an accuracy close to 80%: 0.79 (CI: 0.63-0.90).

Meta-Analysis

Sensitivity: In general, all studies showed very high sensitivity when using a fixed-effects model: 0.94 (CI: 0.93-0.95). Under a random-effects model, sensitivity was slightly lower, with a wider fluctuation range: 0.92 (CI: 0.88-0.95). No significant differences were observed between the index test (FAI) and the comparator test (3Q-TMD) (Figure 2).

Figure 2: Forest plot of sensitivity 

Specificity: TMD screening tests showed lower specificity values than sensitivity, with greater heterogeneity among the results. Using a random-effects model, the meta-analysis indicated a specificity of 0.79 (CI: 0.63-0.90) (Figure 3).

Figure 3: Forest plot of specificity. 

Additional Measures: Diagnostic Odds Ratio (DOR):

The overall DOR values for the set of tests were 3.81 (CI: 3.10-4.53). Based on the table of point estimates for the Diagnostic Odds Ratio, the analyzed tests demonstrated high sensitivity and specificity in terms of diagnostic accuracy, making them effective screening tools (Figure 4).

Figure 4: Forest plot of the Diagnostic Odds Ratio. 

Proportional Hazards Model Approach: The proportional hazards model was used to estimate the performance of the tests on a receiver operating characteristic (ROC) curve. Under the homogeneity model, the overall results for the FAI and 3Q-TMD tests yielded an estimated area under the curve (AUC) of 0.95 (CI: 0.97-0.92). Under the heterogeneity model, the results were slightly lower but still demonstrated very good performance, with an AUC of 0.94 (CI: 0.97-0.91) (Figure 5).

Figure 5: Area under the curve. 

Reference Tests

The RDC/TMD was developed in 1992 for research purposes only. Later, in 2014, the DC/TMD expanded its use to clinical settings. These diagnostic tools are intended to establish reliable, standardized, and validated criteria for diagnosing TMD subtypes, as one of the main methodological issues in correlational research is the precise definition of the criteria applied ¹⁴. Criterion validity for the screening tests was established in relation to the DC-TMD. Although the DC/TMD is reliable and valid, its routine use for clinical triage of TMD is not practical, as its evaluation protocol is time-consuming and requires proper interpretation of its complex algorithms ¹⁷. In turn, screening tests for TMD detection provide a quick and simple way to determine which patients would benefit from a specific diagnosis.

It is worth noting that most of the screening studies in this review used the DC-TMD as the reference test. While the DC-TMD is the updated version of the RDC-TMD criteria, no significant difference in diagnostic accuracy was observed when the studies were analyzed through meta-analysis.

Criterion Validity of the 3Q-TMD Test

The criterion validity of the 3Q-TMD test was established against the DC-TMD reference standard using two patient samples: one from the general population and the other from patients attending a specialized center. When the positive responses to the 3Q-TMD questions were compared with the reference test, a substantial difference between the settings was observed. However, the main reason for this difference may be due to the DC-TMD symptom questionnaire ¹⁶.

Upon examining the results in each setting, two key differences emerged. First, the time frame: although both questionnaires are based on reported symptoms, the 3Q-TMD test covers a one-week period, while the reference test relies on symptoms perceived over the past 30 days. Second, the phrasing of the questions: the first question in the 3Q-TMD questionnaire focuses on reported pain, and the second on functional pain. However, to qualify for a DC pain/TMD diagnosis, the criteria require the pain to be caused or modified by function. The observed differences in the sensitivity and specificity of the 3Q-TMD test may be attributed to these two factors. Nevertheless, the predictive values were high, particularly the negative predictive values, indicating that these questions are excellent for ruling out a TMD diagnosis.

The question related to functional disorders deserves a separate paragraph. There is ongoing discussion about the prognosis of joint sounds and the clinical evaluation possibilities for intra-articular TMD ³⁸. The DC/TMD has shown moderate to poor validity for intra-articular TMD. In this sense, the reference standard would perform better as a screening tool than as a diagnostic instrument ⁽³⁹.

Recently, the validity of the RDC-TMD and DC-TMD clinical protocols for evaluating intra-articular disorders has been questioned ⁴⁰. Imaging studies such as MRI, CT, and even arthrography have been proposed as more accurate alternatives. These methods, however, would only be justified if their results could alter the therapeutic protocol. It is important to note that even in asymptomatic individuals, disc displacements may be observed on MRI in approximately 30% of the population ^41,42. Unlike the two pain questions, the results of the intra-articular disorders question showed varying utility depending on the setting in which it was applied. When applied to the general population, it was useful for ruling out the absence of intra-articular TMD. However, when applied in specialized settings, its high positive predictive value indicated a probable intra-articular TMD. In summary, while the validity of the intra-articular TMD question ranges from fair to moderate, its high specificity makes it very useful for screening, particularly for ruling out dysfunction when the results are negative.

Criterion Validity of the FAI Test

Unlike the 3Q-TMD test, the Fonseca Anamnestic Index identifies individuals with TMD from a score. The FAI was originally designed to determine the severity of symptoms. All FAI studies have demonstrated high accuracy in detecting pain-related and intra-articular TMDs, with this study reporting an area under the curve ranging from 0.93 to 0.98 across various observations ^21,32,36. Studies analyzing the FAI established scores derived from questions that included individuals with pain, those with intra-articular disorders, and the sum of all symptoms. In this regard, the classification was similar to that used in the 3Q-TMD test. The FAI studies, however, differed in their cutoff points, as each study defined different values to achieve the highest accuracy relative to the reference test. These variations may be explained by the heterogeneity of populations and methods used. FAI cutoff points for ruling out individuals without TMD ranged from 0 to 20, with higher scores correlating with increased symptom intensity. While uncertainties remain regarding the utility of assigning a degree of symptom severity, various studies have confirmed the FAI’s ability to identify TMD. Overall, the FAI appears to be highly sensitive for detection, though its specificity is relatively low. The lower specificity observed with the FAI may be attributed to the inclusion of items unrelated to TMD, such as headaches, neck pain, parafunctional habits, malocclusion, and emotional stress. This prompted further investigation into the questionnaire’s dimensionality and psychometric properties ⁽⁴³. The study confirmed the FAI's multidimensionality, identifying a primary five-item dimension, which resulted in the development of the five-question short-form FAI. Research on this abbreviated version showed improved accuracy, with an area under the curve of 0.97 and significantly higher specificity (95.5%) relative to the reference test.

Analysis of the FAI and 3Q-TMD Tests

The results of this study demonstrated that, when comparing both tests, the larger number of published studies on the FAI test allows its diagnostic accuracy results to be considered reliable, despite the significant heterogeneity among studies. In contrast, only two studies were found on the 3Q-TMD test. In this case, the reliability of its results is supported by the low risk of bias assessed in both studies. It is important to note that, despite the heterogeneity in methods and settings, the FAI and 3Q-TMD screening tests have shown very good diagnostic accuracy. Only two studies reported low specificity: Lovgren’s study on the 3Q-TMD test in a specialized setting (specificity 0.34) and Stasiak’s study on the FAI test (specificity 0.26), also conducted in a secondary setting. Regarding Lovgren’s study, performed in a specialized orofacial pain clinic, the high false positive rate could be attributed to the greater frequency of facial pain unrelated to TMD in such settings compared to primary care environments. It is important to emphasize that the DC/TMD was selected due to its reliability and validity for the most common TMD diagnoses. However, numerous additional pain conditions can affect normal jaw function, such as neuropathic pain, atypical odontalgia, fibromyalgia, and cervical pain. The prevalence of both painful and non-painful conditions is expected to be much higher in specialized orofacial pain clinics than in general population-based clinics. Similarly, it is logical that rarer conditions would also have a much higher prevalence in specialized settings ⁴⁴. Therefore, affirmative responses to the 3Q-TMD test in a specialized clinic may be associated with a TMD diagnosis but could also reflect several differential diagnoses. Ultimately, this may explain the increase in false positives and the consequent decrease in specificity.

The same hypothesis could be applied to the results of the Stasiak study. However, there are valid reasons to attribute this difference to the risk of bias. In particular, the Stasiak study does not clearly specify how the reference test was applied nor how patient flow and timing were managed. The reference standard used in these studies relies on a diagnostic system based on strict criteria, with both the clinical history and examination being meticulously structured. The data are subsequently processed using predefined algorithms, resulting in a probable diagnosis. Additionally, the examiner requires appropriate training. Therefore, improper handling of the reference test could explain the discrepancy in results. Supporting this explanation is the study by Yap, which also used the FAI and was conducted in a secondary setting. However, it reported a much higher specificity (0.88). Based on the available information, it is not possible to determine the reasons behind the differences in specificity observed for the FAI test in secondary settings. Overall, these findings highlight the need for further studies of this nature to elucidate such differences. It is important to clarify, however, that when these tests are regarded as screening tools rather than diagnostic instruments, a loss in specificity does not constitute a limitation. In fact, it is expected that screening tests do not exclude any potential patients with the condition, even at the “expense” of increased false positives. In this sense, unlike diagnostic studies, screening studies prioritize sensitivity over specificity, as it is crucial to detect all individuals who may be at risk ⁴⁵.

In terms of sensitivity, the results of almost all studies demonstrated a very good ability to detect TMD, with the exception of the Zagalaz study, which showed a slight decrease (Sensitivity 0.78). However, this result may be questioned, as it originates from the study with the highest risk of bias. Despite this, both the FAI and the 3Q-TMD showed good performance in identifying individuals with TMD.

While there may be uncertainties regarding specific aspects of the results, the lack of well-established tools for TMD detection underscores that this evidence, although not definitive, represents the best currently available and supports promoting the use of these instruments. Since the 3Q-TMD test has been validated in its original language and the FAI in only three other languages, researchers should be encouraged to validate either of these tools in their respective languages.

This systematic review has some limitations. First, only a small number of studies were identified, with variable results regarding the diagnostic accuracy of each test and significant methodological heterogeneity. Second, a high risk of bias was observed.

Future guidance

As a guideline for future studies, it is recommended that researchers take special care in patient selection. Specifically, the inclusion of patients as "cases and controls" should be avoided, as this may bias the results, according to specialists in diagnostic accuracy testing. Notably, among all the studies analyzed, this was the aspect that had the greatest influence on the risk of bias assessment. The heterogeneity and small number of studies found indicate that TMD screening tests are only beginning to be recognized as necessary tools in the field of craniofacial pain of non-odontogenic origin. This observation is significant in this review because, despite some uncertainties in the results, the lack of validated tools for TMD detection highlights that this evidence, while not conclusive, represents the best currently available and supports promoting their use. Given that these instruments have been validated in their original language and, in the case of the FAI, in only three other languages, researchers are encouraged to validate them in their own language.