Based on the accumulating knowledge of the pathogenesis and diagnosis of persistent somatic symptoms, the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-5) introduced major changes in the way of defining somatoform disorders in 2013 (1). The former category, “Somatoform and Related Disorders,” was replaced by a new category, i.e., “Somatic Symptom and Related Disorders” (SSD). While a medical explanation for the occurrence of somatic symptoms was an exclusion criterion for somatoform disorders in DSM-IV (2), SSD is characterized by somatic symptoms that are distressing or result in a significant disruption of daily life, regardless of their etiology (A criterion), by excessive thoughts, feelings, or behaviors regarding the somatic complaints or associated health concerns (B criterion), as well as by a persistent state (at least 6 months) of being symptomatic (C criterion). In 2022, the International Classification of Diseases, 11th Revision (ICD-11) introduced a similar new diagnostic category, i.e., Bodily Distress Disorder (3).

Given the high prevalence of somatoform disorders and medically unexplained symptoms in primary care (4), the shift to SSD is of major relevance to general practitioners. In a recent large survey, German general practitioners estimated the proportion of their patients fulfilling all the criteria of SSD to be 7.7% (5). Yet, the prevalence of SSD in primary care is still unclear as rigorous studies using semi-structured clinical interviews to make the diagnosis are lacking in this setting. Many studies estimate the prevalence of SSD using proxy diagnoses operationalized by a combination of self-report questionnaires (6).

The Patient Health Questionnaire-15 (PHQ-15) (7) and the Somatic Symptom Disorder—B Criteria Scale (SSD-12) (8, 9) are standardized, validated, and freely available patient-reported tools that can assist in assessing the A and B criteria of SSD when used in combination. Two recent studies comparing the combined results of the PHQ-15 and the SSD-12, with semi-structured diagnostic interviews for SSD as the reference standard, found good diagnostic accuracy (area under the receiver operating characteristic curve 0.77 and 0.84, respectively) (10, 11). However, study participants were recruited in a psychosomatic outpatient clinic in Germany (10) and in a variety of hospital outpatient clinics in China (11), where the prevalence of SSD was high (56 and 34%). Furthermore, the combination of recommended cut-off values and reported estimates of sensitivity and specificity differed considerably between the two studies.

In 2016, we performed a cross-sectional study in a primary care population with the primary aim of investigating differences in socio-demographic, somatic, and psychological characteristics between patients consulting their general practitioner either with or without an appointment (12). Furthermore, we investigated the psychometric characteristics and validity of the SSD-12 in primary care (9). However, as the study did not comprise a (semi-)structured diagnostic interview as the reference standard, we could neither directly investigate the prevalence of SSD nor optimal cut-offs and diagnostic accuracy. As recommendations for cut-off values have now been published (10, 11) we performed a secondary analysis of our data. Originally, we aimed to investigate differences in characteristics between patients who tested positive and negative for possible SSD. However, the two different cut-off recommendations resulted in highly discrepant proportions of test positives, which seemed hardly compatible with the sensitivity and specificity estimates reported by Toussaint et al. (10) and Cao et al. (11). In this manuscript, we describe and interpret our findings and discuss the implications of the found inconsistencies for future use of screening tools for SSD in primary care.

Applying the cut-off value combinations recommended by the published diagnostic accuracy studies performed in high-prevalence settings of SSD to our unselected primary care patients resulted in two very different proportions of participants who tested positive in SSD by the self-report questionnaires PHQ-15 and SSD-12 (9 and 23%, respectively). The sensitivity and specificity estimates found by Toussaint et al. for combination #1 were incompatible with our findings, and the estimates found by Cao et al. for combination #2 were partly compatible.

Our study was not a diagnostic accuracy study comparing questionnaire findings with diagnoses based on validated semi-structured or structured interviews as reference standards. Therefore, the true prevalence of SSD among our study participants remains unknown, and we could not directly investigate sensitivity and specificity for defined cut-off values. However, rigorous prevalence and diagnostic accuracy studies for SSD in primary care are still lacking. Recent scoping reviews (6, 13) found that available prevalence estimates for SSD in primary care or the general population are based on proxy diagnoses based on self-report questionnaires. Other studies [e.g. (14)] also use questionnaire combinations to split participants into those with and without suspected SSD. In this situation, the strong inconsistencies in cut-offs identified in our analyzes can provide some important messages to researchers and healthcare professionals.

First, our analyzes strongly suggest that the sensitivity and specificity found by Toussaint et al. (10) for their recommended cut-off combination to detect SSD in psychosomatic outpatients are not applicable to unselected primary care patients. It seems likely that in such a setting, specificity is much higher and sensitivity lower. A higher specificity not only means a higher probability that a person testing negative does not have SSD but also that a positive result makes the presence of SSD more likely compared to a lower value. Lower sensitivity means that fewer persons who actually have SSD will be tested positive, and more patients will be false negative. It is sometimes thought that sensitivity and specificity are “stable” characteristics of a test (15). However, there are many examples showing that these parameters vary between different populations (16–20). The term ‘spectrum effect’ is used if true differences in the distribution of disease (severity) or prior diagnostic testing are the reasons for variations in diagnostic accuracy in different populations (17). It has been shown that specificity can increase while sensitivity decreases if many of the study participants score far below the applied cut-off values (18). On the other hand, sensitivity can increase while specificity decreases if previous diagnostic investigations lead to a selection of patients with more severe symptoms (21, 22). Our findings suggest that the spectrum effect might be large in the case of SSD when comparing psychosomatic outpatients and unselected primary care patients.

Second, our findings also suggest that the cut-off combination #1 is too strict to adequately balance sensitivity and specificity values in unselected primary care patients. An SSD-12 score of ≥23 makes a true diagnosis of SSD very likely (when assuming a true prevalence between 10 and 25%), but many individuals with SSD that score below 23 may be missed. Whether combination #2 with much lower cut-offs is a good choice for primary care remains unclear. In a recent re-analysis of their data (22), Cao et al. discuss a slightly higher cut-off (≥16) when applying the SSD-12 only, without the PHQ-15.

Third, our analyzes show that it is almost impossible at the current state to assess whether prevalence estimates for SSD in primary care based on proxy diagnoses from self-report questionnaires are accurate. There is a possibility of (gross) overestimates or underestimates. Given the example of major depression, an individual patient data meta-analysis of 44 studies on diagnostic accuracy found that the widely used Patient Health Questionnaire-9 (PHQ-9) identifies 2.5 times as many major depression cases as a validated semi-structured diagnostic interview (23). Bayesian approaches have been proposed to estimate prevalence, sensitivity, and specificity in studies without diagnostic interviews, taking into account prior information from diagnostic studies (24). Recently, a Bayesian latent class model was used to account for imperfect diagnostic accuracy in the interpretation of results from a large survey in 27 European countries using the PHQ-8 (a shortened version of the PHQ-9). Based on the PHQ-8 alone, this study estimated an overall prevalence of major depression of 6.4%. Taking into account the diagnostic misclassification, the study found a much lower prevalence of 2.1% (25). This could be a similar problem to our hypothesis regarding a misjudgment of SSD. However, as only two diagnostic studies with two very different cut-off recommendations are available, it may be too early to conduct such analyzes to estimate the prevalence of SSD. Our approach is descriptive, non-inferential, and easily applicable for researchers who want to roughly check whether the results of their questionnaire survey are consistent with published estimates of diagnostic accuracy. However, when researchers apply our approach, they must be aware that both sensitivity, specificity and estimated expected frequencies are all subject to uncertainty.

Despite the different cut-off combinations recommended by Toussaint et al. (10) and Cao et al. (11), the high estimates found for the area under the curve in both studies suggest that the combination of PHQ-15 and SSD-12 show good diagnostic accuracy in terms of detecting patients at risk for SSD. Particularly in primary care, it also could be considered using the SSD-12 alone. The area under the curves in available analyzes (11, 12) suggests satisfactory to good accuracy. Furthermore, in a recent study of primary care patients, SSS-8 scores (a shorter version of the PHQ-15) were found to be more influenced by symptoms of uncomplicated acute infections than SSD-12 scores (26). In any case, setting specific cut-off values for self-report questionnaires covering the A and B criteria of the SSD may be needed, especially for research purposes. For clinical use, it is probably more useful to interpret actual questionnaire score values together with other clinical information available instead of dichotomizing the questionnaire findings into SSD positive and SSD negative. In general, the PHQ-15 and SSD-12 values can be valuable for general practitioners to identify patients who need psychological help due to strong concerns about their persistent somatic symptoms. Obviously, the PHQ-15 and SSD-12 alone are not sufficient to make a diagnosis of SSD, but they give solid indications of the presence of SSD.

Our study was conducted in a limited number of practices in one region of Bavaria. As the primary aim of the original study was to investigate differences between patients with and without appointments, our study sample is probably not fully representative of all patients seeking care in the study practices. Patients without appointments were slightly younger and more often had acute complaints than patients with appointments. However, results regarding SSD were almost identical in both groups. Furthermore, our case mix and mean age compare well with a large study of consecutive primary care patients in Germany (27). Therefore, although our study participants may not be fully representative of unselected primary care patients in Germany, it seems very unlikely that our results are subject to major bias regarding the applicability of the recommended cut-off scores.

Our analyzes strongly suggest that the sensitivity and specificity found by Toussaint et al. (10) for their recommended cut-off combination #1 in psychosomatic outpatients are not applicable to unselected primary care patients and that the cut-off for the SSD-12 (≥23) seems too high. The cut-off combination #2 recommended by Cao et al. (11) seems more applicable. Prevalence estimates for SSD based on proxy diagnoses from self-report questionnaires using cut-off recommendations derived from other clinical settings have a risk of resulting in incorrect classifications in primary care. Studies that compare screening findings by questionnaires with validated diagnostic interviews as reference standards in primary care populations are urgently needed in order to make statements on the accuracy of the scales to detect patients at risk for SSD.

The original contributions presented in the study are included in the article/Supplementary material. Further inquiries can be directed to the corresponding author.

The studies involving humans were approved by Ethikkommission Technical University Munich. The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required from the participants or the participants’ legal guardians/next of kin because due to anonymised data collection, oral consent was obtained.

VS: Conceptualization, Data curation, Formal analysis, Writing – original draft, Writing – review & editing. AT: Supervision, Writing – review & editing. AH: Conceptualization, Formal analysis, Methodology, Writing – review & editing, Writing – original draft. CT: Writing – review & editing. BR: Project administration, Writing – review & editing. PH: Supervision, Writing – review & editing. JG: Supervision, Writing – review & editing. AS: Funding acquisition, Methodology, Project administration, Supervision, Writing – review & editing. KL: Conceptualization, Investigation, Methodology, Supervision, Writing – original draft, Writing – review & editing.

The POKAL-Study-Group [PrädiktOren und Klinische Ergebnisse bei depressiven ErkrAnkungen in der hausärztLichen Versorgung (POKAL, DFG-GRK 2621)] consists of the following principal investigators: Tobias Dreischulte, Peter Falkai, Jochen Gensichen, Peter Henningsen, Markus Bühner, Caroline Jung-Sievers, Helmut Krcmar, Karoline Lukaschek, Gabriele Pitschel-Walz, and Antonius Schneider. The following doctoral students are also members of the POKAL-Study-Group: Jochen Vukas, Puya Younesi, Feyza Gökce, Victoria von Schrottenberg, Petra Schönweger, Hannah Schillock, Jonas Raub, Philipp Reindl-Spanner, Lisa Hattenkofer, Lukas Kaupe, Carolin Haas, Julia Eder, Vita Brisnik, Constantin Brand, Chris Ebert, Marie Emilia Vogel, Clara Teusen, and Katharina Biersack.