Response to the Letter to the Editor Regarding “Knowledge and Attitudes Toward Attention-Deficit/Hyperactivity Disorder Among Physicians and Senior Medical Students in Jordan” [Response To Letter] Authors: Redab Al-Ghawanmeh, Mohanad Odeh, Tamara Kufoof, Mohammad Al-Qudah, Mohammad Hassan Al-thnaibat, Salma Burayzat, Sara Suwan, Hayfaa A Y Alhazami Published: 2026 Dear editor We thank Rana and Maurya for their thoughtful engagement with our article, “Knowledge and Attitudes Toward Attention-Deficit/Hyperactivity Disorder Among Physicians and Senior Medical Students in Jordan”.1 Post-publication correspondence provides a valuable opportunity to clarify methodological decisions that were necessarily reported concisely in the original article. We welcome the opportunity to address the points raised and to place them within the intended scope of our applied, cross-sectional medical-education study. Questionnaire Validation and Psychometric Considerations Rana and Maurya raise important questions regarding structural validity, test–retest reliability, and details of the content-validation process. Additional psychometric analyses can provide further evidence regarding an instrument’s measurement properties. However, our study did not aim to develop a new ADHD knowledge or attitude scale. Rather, the questionnaire was adapted from previously validated instruments and aligned with DSM-5 diagnostic criteria for application within the Jordanian medical and educational context.1 Accordingly, validation focused on content validity, contextual relevance, and internal consistency. Content validation involved a panel of 10 experts from health-related academic and clinical fields relevant to ADHD, medical education, public health, and health-professions education in Jordan. Item-level validity was assessed using Lawshe’s Content Validity Ratio (CVR), and section-level validity was assessed using the Content Validity Index (CVI). Items meeting the predefined CVR criterion were retained in accordance with established methodological guidance.2,3 Internal consistency was also formally assessed: Cronbach’s alpha coefficients were 0.77 for the knowledge domain, 0.71 for the attitude domain, and 0.78 for the overall questionnaire.1 Thus, while structural validity and test–retest reliability would provide additional psychometric information, their absence should be interpreted in the context of an adapted instrument used for an applied cross-sectional investigation rather than a primary scale-development study. Rationale for the Three-Point Scoring System The three-point scoring system was predefined to distinguish three educationally meaningful response states: correct knowledge (+1), acknowledged uncertainty (“not sure” = 0), and incorrect knowledge or misconception (−1). This distinction is relevant in health-professions education because acknowledged uncertainty and an incorrect belief are not necessarily equivalent educational states. Related assessment approaches have used positive credit for correct responses, no penalty for acknowledged uncertainty, and negative marking for incorrect responses to distinguish lack of knowledge from incorrect knowledge and reduce random guessing.4,5 Similar three-state scoring logic has also been applied in recent health-professions education research.6 For completeness, the predefined knowledge thresholds were >12 for good knowledge, 1–12 for fair knowledge, and ≤0 for poor knowledge. For attitudes, positive responses were scored +1, negative responses −1, and “not sure” responses 0; an overall score >0 represented a positive attitude and ≤0 a negative attitude. These numerical thresholds were abbreviated in the published article as “predefined cutoffs”; we provide them here to make the scoring framework explicit for readers. Variable Selection and Multivariable Modeling The Letter also raises the use of bivariate associations to identify candidate variables for multivariable modeling. We acknowledge that variable-selection strategies have limitations and that selection based on bivariate significance is not intended to substitute for theory-based confounder selection when causal inference is the objective. Variable-selection approaches vary according to whether the purpose of a model is causal inference, prediction, confounder control, or exploratory explanation.7,8 Our regression analyses were exploratory and explanatory rather than causal or predictive. Bivariate screening was therefore used as a pragmatic and parsimonious strategy for examining whether observed associations persisted after adjustment within the available dataset. The resulting adjusted estimates should accordingly be interpreted as statistical associations within this cross-sectional sample, rather than as causal effects or coefficients from a clinical prediction model. Ordinal Regression and the Proportional-Odds Assumption The Letter highlights the significant test of parallel lines in the student model. Importantly, this diagnostic finding was explicitly evaluated and reported in the original article. As stated in the original article, the test of parallel lines was significant (χ2(4)=20.726, p<0.001), indicating that the proportional-odds assumption was not fully supported; estimates from the student model were therefore explicitly identified as requiring cautious interpretation.1 Ordinal logistic regression was selected because knowledge was classified into three inherently ordered educational levels—poor, fair, and good—and the cumulative-logit model preserves this ordering.9 Although multinomial logistic regression does not require the proportional-odds assumption, it treats outcome categories as nominal rather than ordered and was therefore not considered the preferred alternative for this inherently ordinal outcome. Partial proportional-odds and generalized ordered-logit models are alternative approaches that retain the ordinal structure when the proportional-odds assumption is not satisfied.10 Accordingly, the adjusted common odds ratio from the student model should not be interpreted as demonstrating a uniform effect across all knowledge thresholds. Importantly, however, the observed association between pediatric rotation and knowledge did not depend solely on this adjusted estimate. Among students who had completed a pediatric rotation, 76.4% demonstrated good ADHD-related knowledge compared with 26.0% among those who had not completed the rotation (p<0.001; Cramér’s V=0.36).1 In the adjusted model, pediatric rotation remained associated with higher knowledge (OR=3.370, 95% CI 1.075–10.561; p=0.037), subject to the aforementioned limitation.1 We therefore regard this finding as an educationally relevant, hypothesis-generating association rather than evidence of a causal effect. Physician Subgroup Analysis and Precision The Letter also raises concerns regarding small physician subgroups. It is important to distinguish categories presented descriptively from variables entered into the regression model. Small specialty categories, including psychiatry and obstetrics and gynecology, were reported to characterize the physician sample but were not entered as specialty predictors in the physician ordinal regression model.1 The association between having children and knowledge was accompanied by a wide confidence interval (OR=3.223, 95% CI 1.003–10.363; p=0.049) and should therefore be interpreted with appropriate caution.1,11 The physician findings, however, did not depend on this association alone. Working in primary healthcare rather than a central hospital was associated with lower odds of higher knowledge (OR=0.418, 95% CI 0.239–0.729; p=0.002), while personal ADHD experience was associated with higher knowledge (OR=2.508, 95% CI 1.352–4.652; p=0.004).1 The Discussion therefore interpreted these findings collectively as associations with experiential and contextual factors rather than basing its conclusions on parental status or any single predictor. Multiple Bivariate Comparisons The Letter also raises the issue of multiple testing when interpreting individual p-values. Formal multiplicity adjustment was not applied because the bivariate analyses were exploratory rather than a prespecified family of confirmatory hypothesis tests. The use of multiplicity adjustment in exploratory analyses remains methodologically debated, particularly because reducing type I error may increase type II error.12 Importantly, interpretation was not based on statistical significance alone. Cramér’s V effect sizes accompanied the bivariate analyses,1 and the study’s conclusions were based on the combined pattern of descriptive findings, effect sizes, and adjusted analyses rather than isolated p-values. Multiplicity-control procedures remain reasonable for future studies designed around prespecified confirmatory hypotheses. Precision of the Principal Descriptive Findings To further characterize the precision of the principal descriptive findings, 305 of 553 participants demonstrated good ADHD-related knowledge, corresponding to 55.2% (95% CI 51.0–59.3%), while 292 of 553 demonstrated positive attitudes, corresponding to 52.8% (95% CI 48.6–57.0%). Reporting confidence intervals alongside proportions provides additional information regarding the precision of these estimates, consistent with established recommendations for interval estimation of binomial proportions13 and STROBE guidance on reporting measures of precision for principal estimates.14 Interpretation of Pediatric Clinical Exposure Finally, the interpretation of the pediatric-rotation finding should be considered within the cross-sectional design of the study. As clearly reflected in the published article, the observed association was not presented as evidence of causation.1 The Discussion suggested that pediatric clerkships may provide meaningful exposure to developmental and behavioral concerns and explicitly identified the cross-sectional design as a limitation.1 Accordingly, the pediatric-rotation finding should be understood as an educationally relevant, hypothesis-generating association. The proportional-odds limitation appropriately constrains interpretation of the adjusted student estimate but does not negate the observed distributional pattern, bivariate association, or effect size. Longitudinal and interventional studies are required to determine whether structured clinical or educational exposure produces improvements in ADHD-related knowledge or professional attitudes. Conclusion We thank Rana and Maurya for their constructive engagement with our work. Their comments provide an opportunity to clarify aspects of questionnaire validation, scoring, statistical analysis, and precision that were reported concisely in the original article. As reported in the original article, the proportional-odds diagnostic in the student model and the precision of individual estimates were taken into account in interpreting the findings. At the same time, these considerations do not alter the principal descriptive finding of a meaningful knowledge–attitude gap among physicians and senior medical students or the value of the observed educational and experiential associations as hypotheses for future investigation. Disclosure The authors report no conflicts of interest in this communication. References