Development and psychometric validation of the 13-item physician health and quality of life (Afya MedQoL) index: a nationwide cross-sectional survey of Brazilian physicians Authors: Marcelo Gobbo Jr, Marina Silveira de Resende, Isabelle Martins Santos, Eduardo Cardoso de Moura, Renata de Almeida Pedro Published: 2025 ABSTRACT AbstractObjectivesTo develop and validate a concise, physician-specific quality-of-life scale (Afya MedQol) Index, and to examine sociodemographic and occupational correlates of physician well-being in Brazil.DesignNation-wide, cross-sectional, web-based survey employing split-sample psychometric validation: exploratory factor analysis, confirmatory factor analysis (CFA), multidimensional graded-response item-response theory calibration and post-stratification weighting to the 2025 Brazilian medical demographic census.SettingPhysicians practising in all five Brazilian macro-regions and 27 state medical councils (2 July 2024–6 August 2024).ParticipantsA total of 2005 fully responding physicians (56% women; mean age 38.5±12.6 years).Primary outcome measuresDomain scores—F1: Quality of Life, F2: Institutional Support, F3: Perceived Stress—and a Fisher-information-weighted global score. Convergent validity with WHOQOL-Bref; subgroup differences (Cohen’s d, partial η²).ResultsCFA supported a three-factor, 13-item solution (Comparative Fit Index=0.987; Root Mean Square Error of Approximation=0.050) with strict invariance across sex and graduation-year quartiles. Item-response modelling showed marginal reliability ρ=0.82. Convergent validity with WHOQOL-Bref domains was strong (r=0.55–0.73). Weighted mean (95% CI) scores were: Global 67.2 (66.5–67.9), F1 69.5 (68.9–70.2), F2 64.1 (63.4–64.8) and F3 62.5 (61.8–63.2).Women scored higher on Perceived Stress (F3) than men by 5.6 points (weighted mean difference 5.6, 95% CI 4.0 to 7.2; Cohen’s d 0.28). Physicians working ≥60 hours/week scored higher on Perceived Stress than those working ≤44 hours/week by 8.8 points (weighted mean difference 8.8, 95% CI 6.7 to 10.9; Cohen’s d 0.40). Well-being increased with income up to ~BRL 25 000 per month and plateaued thereafter.ConclusionsAfya MedQoL Index is a psychometrically robust, 13-item instrument capturing personal, organisational and stress dimensions of physician well-being. Perceived Stress—shaped by workload, institutional climate and gender—was the most discriminative domain, while additional income beyond an upper-middle-class threshold conferred negligible benefit. Findings support policies aimed at capping duty hours, fostering ‘just-culture’ environments and expanding flexible work models. STRENGTHS AND LIMITATIONS OF THIS STUDY Robust psychometric validation: a split-sample strategy (exploratory+confirmatory factor analyses) followed by multidimensional graded-response Item Response Theory calibration provides rigorous evidence for the structural validity of the 13-item Afya MedQoL Index. Robust psychometric validation: a split-sample strategy (exploratory+confirmatory factor analyses) followed by multidimensional graded-response Item Response Theory calibration provides rigorous evidence for the structural validity of the 13-item Afya MedQoL Index. National representativeness: post-stratification weighting aligns the sample with the 2025 Brazilian Medical Census, enhancing generalisability of prevalence estimates. National representativeness: post-stratification weighting aligns the sample with the 2025 Brazilian Medical Census, enhancing generalisability of prevalence estimates. Convergent validity: high correlations with the WHOQOL-Bref domains (|r|≥0.55) support the construct validity of the new instrument. Convergent validity: high correlations with the WHOQOL-Bref domains (|r|≥0.55) support the construct validity of the new instrument. Design constraints: the cross-sectional, volunteer, web-based design precludes causal inference and may under-represent physicians lacking reliable internet access or experiencing severe distress. Design constraints: the cross-sectional, volunteer, web-based design precludes causal inference and may under-represent physicians lacking reliable internet access or experiencing severe distress. Content and measurement gaps: the scale omits positive-psychology constructs such as purpose and flow and relies exclusively on self-report; longitudinal studies incorporating objective or behavioural markers are needed. Content and measurement gaps: the scale omits positive-psychology constructs such as purpose and flow and relies exclusively on self-report; longitudinal studies incorporating objective or behavioural markers are needed. Introduction Physicians are exposed to a unique constellation of occupational hazards—prolonged working hours, high cognitive load, emotional labour and frequent exposure to suffering—that erode well-being and heighten the risk of burnout, depression, substance misuse and early exit from the workforce. Meta-analyses estimate global burnout prevalence among physicians between 40% and 60%,14 with downstream consequences that include a two to threefold increase in major medical errors and a 17 % rise in turnover intentions.1 2 5 6 In Brazil, these pressures intersect with rapid expansion of medical schools and widening regional inequities in resource allocation: the 2025 National Medical Demography Census estimates 635,706 licensed physicians in 2025 across the country, yet persistent ‘medical deserts’ remain in the North and Northeast, while a feminising and progressively younger workforce faces escalating workload demands.7 Over the past two decades, the concept of physician quality of life (QoL) has expanded from a narrow focus on burnout to a multidimensional paradigm that integrates mental, physical, social and occupational domains.8 Positive psychology concepts—purpose, engagement and resilience—now complement traditional negative indicators such as emotional exhaustion or secondary trauma.9 10 This holistic view is supported by job demands-resources theory and self-determination theory, both of which highlight the interplay between individual resources and organisational context in shaping professional.11 12 Despite conceptual advances, measurement approaches have lagged behind. Systematic reviews identify more than 60 instruments applied to physicians, yet most were adapted from generic populations, emphasise isolated constructs (eg, burnout), or lack of rigorous psychometric validation.13 14 Only a handful—such as the MAPPWrQoL15 or the Vietnamese ProQoL16—are occupation specific, culturally tailored and multidimensional. Critical methodological gaps persist: (1) Inconsistent definitions of physician well-being, (2) Small, convenience samples, (3) Cross-sectional designs precluding responsiveness testing and (4) Underrepresentation of systemic workplace stressors.9 17 High-quality measurement is the cornerstone of effective policy. In the absence of a concise, culturally attuned and psychometrically rigorous instrument, it is impossible to (1) Quantify the burden of poor well-being in Brazilian physicians, (2) Benchmark institutions or (3) Evaluate interventions such as duty hour limits or peer support programmes. Item Response Theory (IRT) offers clear advantages over classical scoring—trait-level precision, information-weighted items and invariant measurement—but has seldom been applied to physician well-being scales.18 Addressing these gaps will generate a tool that is both scientifically robust and operationally feasible for large scale surveillance. In Brazil, most studies employ generic instruments such as the WHOQOL-Bref or the Maslach Burnout Inventory, which capture limited facets of the physician experience and show modest discriminative power across specialties, contract types and income strata.1922 No brief, psychometrically sound scale exists that simultaneously assesses personal QoL, institutional support and perceived stress. The present study aimed to develop, calibrate and validate the Afya Physician Health and Quality of Life (Afya MedQoL) Index in a nationwide sample of Brazilian physicians. Specific objectives were (1) To construct a parsimonious item bank that captures three theorised domains—QoL, Institutional Support and Perceived Stress—using Exploratory Factor Analyses (EFA) and Confirmatory Factor Analyses (CFA), (2) To estimate item parameters with a multidimensional Graded Response Model (GRM) and derive information-weighted global and domain scores, (3) to establish reliability, convergent validity (against the WHOQOL-Bref) and measurement invariance across sex, career stage and income strata and (4) Quantify the association between Afya MedQoL Index scores and key sociodemographic, occupational and psychosocial determinants to inform targeted policy interventions. Materials and methods Design, setting and participants A nationwide, cross-sectional, web-based survey was administered between 2 July 2024 and 6 August 2024 to licensed physicians practising in Brazil. Eligibility required an active registration in one of the 27 Regional Medical Councils and prior use of, or subscription to, a national digital health platform. Invitation links were distributed via email, in-app push notifications and social media posts. Data were collected on an ISO 27001-certified QuestionPro server. Respondents reporting non-medical professions or declining consent were excluded. Participants were recruited through a national web-based platform and professional mailing lists, which may systematically under-represent physicians working in remote or rural areas with limited internet connectivity. To quantify this bias, we compared the unweighted sample distribution with the 2025 Brazilian Medical Census7 across (1) Macro-region, (2) Sex, (3) Age band and (4) Medical speciality (an important characteristic not used in the weighting algorithm). The largest deviation was observed in Family Medicine (sample=4.8% vs census=9.4%), whereas Surgical specialties were slightly over-represented (sample=18.2% vs 14.5%). These discrepancies are partially mitigated by the post-stratification weights applied (see below); nevertheless, caution is advised when extrapolating prevalence estimates to under-sampled sub-populations of rural clinicians. Future studies employing stratified probability sampling or hybrid online/offline recruitment are warranted. Post-stratification weighting To enhance population representativeness, post-stratification (raking) aligned the sample with the 2025 Brazilian medical census margins for sex (50.9% female/49.1% male), macroregion (North 6.5%, Northeast 24.4%, Centre-West 12.1%, Southeast 45.7%, South 11.4%) and age (5-year bands).7 Iterative proportional fitting was conducted using R V.4.3.3 (survey package), achieving convergence after six iterations (tolerance 10⁻⁵). The resulting weighting design effect was 1.31. To manage variance inflation, design weights exceeding four were conservatively trimmed to this threshold. This value, while not universally standardised, aligns with prior recommendations to truncate or smooth extreme weights to reduce sampling variance and instability of estimates.2325 This approach reduced the coefficient of variation of weights from 76% to 49%, decreased the average linearised variance of prevalence estimates by 18%, and minimally impacted point estimates (<0.3 percentage points). All analyses presented herein use trimmed weights. Instrument development Because no single scale simultaneously covers physicians’ general QoL, organisational climate and stress, a bespoke item bank was constructed in four steps:26 27 Concept mapping: three domains were mapped: Quality of Life (WHOQOL-Bref),2830 Institutional Support (job demand–control and effort–reward models)31 and Perceived Stress.32 Concept mapping: three domains were mapped: Quality of Life (WHOQOL-Bref),2830 Institutional Support (job demand–control and effort–reward models)31 and Perceived Stress.32 Item generation: two senior physicians drafted items mirroring the source instruments. Item generation: two senior physicians drafted items mirroring the source instruments. Expert review: a psychologist and two external researchers iteratively refined items in two Delphi rounds.31 33 Expert review: a psychologist and two external researchers iteratively refined items in two Delphi rounds.31 33 Content validity: all 40 candidate items achieved content validity. Content validity: all 40 candidate items achieved content validity. Items used five-point Likert-type responses, and higher scores indicated better status. Sample size, normality and split-half validation With 40 items, the final sample (n=2005) satisfies the recommended 10:1 respondent-to-item ratio.34 The dataset was randomly split into EFA (n=1002) and CFA (n=1003) subsamples, allowing genuine cross-validation and subsequent measurement invariance testing.35 36 For each variable, two complementary tests were applied: the Shapiro–Wilk test, given its good power in small to moderate samples and consistent performance across forms of non-normality,37 38 and the D’Agostino–Pearson K² test, which combines skewness and kurtosis and performs well in medium to large samples.39 Unbiased estimates of skewness and excess kurtosis were reported as well. Multivariate normality was assessed using Mardia’s measures: b1,p for skewness, with statistic n⋅b1,p/6 referenced to a chi-square distribution with df=pp+1p+2/6, and b2,p for kurtosis, with a z-score relative to the theoretical expectation pp+2 under normality.40 41 Missing data In the raw dataset (n=2005) the sentinel value nine indicated a skipped answer. All ‘9’ responses were re-coded to NA. The proportion of missing responses per item ranged from 0.0% (nine items) to 9.5% for the four ‘institutional support’ items, with an overall missingness of 2.9% across the 13 items retained in the final model (online supplemental Table S12). Because item-wise missingness never exceeded 10%, we applied median imputation column-wise before all multivariate analyses. Median imputation is robust for ordinal Likert data and preserves variance better than mean substitution.40 Exploratory factor analysis Assumptions of normality were violated; therefore, a polychoric correlation matrix was analysed.41 Sampling adequacy was acceptable (KMO=0.8095), and Bartlett’s test of sphericity was significant (p<0.001), supporting the factorability of the correlation matrix.42 43 Extraction used principal axis factoring. For an initial check of simple structure, varimax rotation was applied, followed by oblique promax rotation with k=4 to allow factor correlations and improve interpretability.44 The number of factors was determined with Horn’s parallel analysis,45 comparing observed eigenvalues with those expected from random data, and selecting the most parsimonious solution that fit the theoretical model and showed a simple, interpretable structure. A three-factor solution was retained. Items with primary loadings ≥0.60 and crossloadings <0.30 were retained.44 Loading SEs were estimated via bootstrap resampling with 1000 draws.46 Confirmatory factor analysis The three-factor model was estimated with WLSMV, suitable for ordered data.47 Model fit thresholds followed Hu and Bentler:48 Comparative Fit Index (CFI)/Tucker-Lewis Index (TLI) ≥0.95, Root Mean Square Error of Approximation (RMSEA) ≤0.06, Standardised Root Mean Square Residual (SRMR) ≤0.08. Items with standardised loadings ≥0.60 were retained.49 50 Internal consistency was assessed via Cronbach’s alpha (α) and McDonald’s omega (ω), adopting the threshold of ≥0.70 as indicative of adequate reliability.51 Item response theory (IRT) calibration A multidimensional GRM52 estimated item discrimination (a) and thresholds (b). Expected a posteriori (EAP) θ scores were rescaled to T-scores (M=50, SD=10) and then linearly transformed to 0–100 scale.53 For interpretability, F3 (Perceived Stress) is reported in the natural direction, where higher scores indicate greater stress. When computing the global Afya MedQoL Index, the reverse-coded stress direction is retained so that higher global scores consistently indicate better well-being. Marginal reliabilities ≥0.80 were interpreted as indicating adequate measurement precision.54 Measurement invariance Configural, metric, scalar and strict invariance across sex and graduation year quartiles were tested sequentially. Invariance was accepted when ΔCFI ≤0.010 and/or ΔRMSEA ≤0.0150.55 Differential item functioning (DIF) Differential item functioning (DIF) analyses were conducted using ordinal logistic regression implemented with the lordif package (graded-response parameterisation)56 to examine potential DIF across four sociodemographic variables: macro-region (North, Northeast, Central-West, Southeast, South), race/ethnicity (White, Black, Pardo, Asian, Indigenous), gender (Female, Male) and age quartiles. DIF was considered practically relevant if the Bonferroni-adjusted McFadden’s ΔR² exceeded the criterion of 0.02, following standard guidelines for meaningful DIF.56 57 Computation of Afya MedQoL scores After CFA confirmation, the 13 retained items were recalibrated in the three-dimensional GRM. Participant level F1, F2 and F3 θ scores were combined into a global index using information-weights averages, with weights incorporating item discrimination (a), where ajf is item discrimination.18 54 Composite scores were transformed into T-scores and subsequently to a 0–100 centesimal scale to enhance interpretability, following established practices in psychometric literature.58 Across all metrics (θ, T-score and 0–100), higher global index scores indicate better well-being. This unidirectional interpretation also applies to F1 and F2, for which higher values reflect better status in the respective domains. In contrast, when interpreted as a standalone domain, F3 (Perceived Stress) is presented in its natural risk direction, such that higher values indicate greater perceived stress. However, for the computation of the global composite, F3 is reverse-coded so that all components contribute in the same direction and the aggregated instrument remains unidirectional (higher global scores = better well-being). Categories were defined based on θ±1.4 SD intervals, consistent with norm-referenced interpretation framework. Income spline modelling Weighted generalised additive models with cubic regression splines (mgcv, REML) quantified the association between income and each outcome. Monthly income was modelled from 0 to 80 000 BRL (knot spacing 500 BRL); hourly income from 0 to 400 BRL per hour (knot spacing 0.5 BRL). Piecewise linear splines extracted slopes below and above BRL 25 000 per month. Robust sandwich SEs accounted for weighting. Model adequacy was summarised with adjusted R2 and inspection of first derivative plot.59 Visual analytics Gaussian kernel density estimates were computed using Scott’s bandwidth over 400 grid points; densities were normalised to (0, 1)60 and displayed with the viridis palette.61 Convergent validity WHOQOL-Bref scores (0–100) were computed. Convergent validity was tested via Pearson correlations with 95% CIs and computed using Fisher’s z transformation between Afya MedQol metrics and WHOQOL domains; |r|≥0.50 was considered large.62 A sensitivity Spearman ρ analysis produced identical inferences. Subgroup analyses Primary strata comprised sex, age and graduation year quartiles, and income quartiles (monthly and hourly). Secondary and tertiary strata included region, self-declared race/skin colour, marital status, weekly workload (≤44/45–59/60–74/≥75 hour), self-reported mental disorder, negative affect score (1–3 vs 4–5), specialty, workplace type, contract type and work modality. Binary group comparisons were conducted using robust-variance t-tests; ≥3 group comparisons employed Welch ANOVA with Games–Howell post hoc tests, appropriate for heteroscedastic data.63 Effect sizes followed Cohen’s d or partial η²: small (d≥0.20; η² ≥0.01), medium (d≥0.50; η² ≥0.06), large (d≥0.80; η² ≥0.14).62 64 All analyses were conducted using R V.4.3.3. To ensure transparency and reproducibility, the full dataset is publicly available65 and all R code notebooks used in the analyses are provided as supplementary materials. In accordance with best practices, all publicly available data used in the preparation of this article, whether generated by the authors or by other researchers, is properly cited both in the text and in the reference list. Patient and public involvement Patients and members of the public were not involved in the design, conduct, reporting or dissemination plans of this research because the study focused exclusively on licensed physicians. The research question arose from a national professional-development initiative led by the Brazilian Medical Association. Results Of the 2005 physicians who completed the questionnaire, the weighted mean age was 39.7±12.9 years (range 23–94) and 51.2% were women, closely matching census margins after raking (design effect=1.31; table 1). The weighted median year of graduation was 2014 (IQR 2004–2020), indicating that just over half the sample had ≤10 years in practice. Median net income was BRL 18 300 per month (IQR 13 000–28 100) and BRL 95.6 per hour (IQR 70.2–134.9) on the hourly metric. The full speciality breakdown and unweighted descriptives are provided in online supplemental Table S1. Table 1Weighted sociodemographic and economic profile, PHyS QoL scores and work characteristics (n=2005)Domain/variableValue*Age (years)39.7±12.9 (23–94)Female sex1028 (51.2%)Year of graduation2014 (IQR 2004–2020)Net income18 000 BRL per month (IQR 13 000–28 100)Hourly income95.6 BRL per hour (IQR 70.2–134.9)Global QoL (0–100)67.2±15.0F1 Quality of Life69.5±15.4F2 Institutional Support64.1±16.2F3 Perceived Stress62.5±15.8MacroregionNorth 6.4%; Northeast 23.8%; Centre-West 11.9%; Southeast 45.9%; South 12.0%RaceWhite 72.1%; Black/Pardo 24.5%; Asian 2.9%; Indigenous 0.5%Marital statusSingle 40.8%; Married/stable 54.0%; Separated 4.8%; Widowed 0.4%Weekly workload≤44 hour 44.0%; 45–59 hour 20.8%; 60–74 hour 24.4%; ≥75 hour 10.8%Self-reported mental disorder38.6%Recent negative affect (4–5/5)23.9%Predominant workplace†Public hospital 24.6% (ref.); Private hospital 19.1%; Private clinic 26.5%; Mixed 15.6%; Telemedicine 7.6%; Primary care 13.0%*Values are mean±SD (range) or median (IQR) for continuous variables and weighted counts (percentages) for categorical variables.†Percentages exceed 100% because respondents could indicate more than one main setting.QoL, quality of life. Weighted sociodemographic and economic profile, PHyS QoL scores and work characteristics (n=2005) Values are mean±SD (range) or median (IQR) for continuous variables and weighted counts (percentages) for categorical variables. Percentages exceed 100% because respondents could indicate more than one main setting. QoL, quality of life. On the centesimal scale, the weighted global index averaged 67.2±15.0, with domain means of 69.5±15.4 for Quality of Life (F1), 64.1±16.2 for Institutional Support (F2) and 62.5±15.8 for Perceived Stress (F3). Cohen’s d comparing weighted with unweighted means was ≤0.10 for all four scores, that is, well below the threshold for a small effect.62 Given that the 40 items are five-point Likert with limited categories—and to avoid biases from treating thresholds as equal-interval—no assumption of approximate continuity or normality was adopted. In 2005 respondents with no missing data and univariate normality were widely violated: all items failed Shapiro–Wilk (W=0.667–0.909; all p<0.001) and D’Agostino–Pearson K² (9.25–11 357; all p<0.05, most p<0.001), with median skewness −0.17 (−1.35 to 1.64) and median excess kurtosis −0.29 (−1.56 to 2.00). Multivariate normality was likewise rejected by Mardia’s tests (skewness χ² (11 480)=631 293, p<0.001; kurtosis b₂, p=3690.3 vs 1680 expected under H₀, z=776.44, p<0.001). Exploratory factor analysis All Likert-type items departed from univariate and multivariate normality; therefore, a polychoric correlation matrix was analysed.41 Sampling adequacy was acceptable (KMO=0.8095), and Bartlett’s test of sphericity was significant (p<0.001), supporting the factorability of the correlation matrix.42 43 Extraction used principal axis factoring. An initial simple-structure check with varimax rotation was followed by oblique promax rotation with k=4 to allow factor correlations and improve interpretability.44 Horn’s parallel analysis suggested an initial five-factor solution; the most parsimonious and theory-consistent three-factor solution was retained, explaining 46.2% of the common variance (eigenvalues=13.79, 2.53, 2.16).45 After suppressing cross-loadings ≥0.30, 37 items retained primary loadings ≥0.600.45 Internal consistency for the retained item set was excellent (Cronbach’s α=0.93; McDonald’s ω=0.96). The complete pattern and structure matrices with bootstrap SEs from 1000 resamples are reported in online supplemental Table S2.46 Confirmatory factor analysis Two CFA models were tested in the validation half-sample (n=1003) using robust WLSMV estimation.47 Fit criteria met Hu and Bentler’s48 stringent thresholds (CFI/TLI ≥0.95, RMSEA ≤0.06, SRMR ≤0.08). Standardised loadings in Model B ranged from 0.57 to 0.85 (table 2), comfortably above the 0.50 benchmark for practical relevance.24 Composite reliabilities remained satisfactory (α/ω=0.78–0.83). Table 2Goodness-of-fit indices for two confirmatory models (validation subsample, n=1003)ModelItemsχ² (df)CFITLIRMSEA (90% CI)SRMRA – 37-item371 426.3 (591)0.8780.8650.072 (0.068 to 0.076)0.086B – 13-item (trimmed)13214.7 (62)0.9870.9840.050 (0.043 to 0.057)0.047Fit thresholds adopted: CFI/TLI ≥0.95, RMSEA ≤0.06, SRMR ≤0.080.48 Model B meets all criteria, supporting the concise 13-item structure.CFI, Comparative Fit Index; RMSEA, Root Mean Square Error of Approximation; SRMR, Standardised Root Mean Square Residual; TLI, Tucker-Lewis Index. Goodness-of-fit indices for two confirmatory models (validation subsample, n=1003) Fit thresholds adopted: CFI/TLI ≥0.95, RMSEA ≤0.06, SRMR ≤0.080.48 Model B meets all criteria, supporting the concise 13-item structure. CFI, Comparative Fit Index; RMSEA, Root Mean Square Error of Approximation; SRMR, Standardised Root Mean Square Residual; TLI, Tucker-Lewis Index. Model B demonstrated a clear three-factor structure (path diagram on online supplemental figure S1), with moderate positive correlation between F1 and F2 (ρ=+0.29) and inverse correlations with F3 (F1–F3=–0.37; F2–F3=–0.17), all p<0.001. Bootstrap-corrected SEs were ≤0.03 and z-values >22, indicating precise estimation.46 All standardised loadings, SEs and reliability coefficients for the final three-factor model are presented in table 3 and the complete item label is presented in online supplemental Table S3. Table 3Standardised factor loadings (λ) and bootstrap SE for the retained 13-item confirmatory model of the physician health and quality of life indexFactor/item (abbreviated)λSEF1 Quality of LifeLeisure opportunities0.8100.022Enjoyment of life0.7610.023Financial sufficiency0.6480.026Access to information0.6120.030Mobility past 2 weeks0.6090.030Access to health services0.5690.027F2 Institutional SupportEducational handling of errors0.8140.023Coworker support network0.7410.026Mental health support0.7250.029Technical training0.7140.028F3 Perceived StressStress hurts performance0.8470.026Stress hurts relationships0.8210.026Stress led to errors0.6790.026All loadings are statistically significant (p<0.001) and meet or exceed the recommended 0.50 threshold for practical relevance.24 Factors: F1=Quality of Life, F2=Institutional Support, F3=Perceived Stress.SE, bootstrap SE (1000 resamples); λ, standardised CFA loading. Standardised factor loadings (λ) and bootstrap SE for the retained 13-item confirmatory model of the physician health and quality of life index All loadings are statistically significant (p<0.001) and meet or exceed the recommended 0.50 threshold for practical relevance.24 Factors: F1=Quality of Life, F2=Institutional Support, F3=Perceived Stress. SE, bootstrap SE (1000 resamples); λ, standardised CFA loading. Collectively, the trimmed 13-item solution exhibits close global fit, strong convergent loadings, acceptable discriminant patterning and robust composite reliability, satisfying contemporary recommendations for scale refinement and validation.26 27 Item response theory calibration The 13 items retained in Model B were recalibrated with a three dimensional GRM52 after reverse scoring the three stress indicators so that higher categories uniformly reflected better status. Convergence was achieved in 22 EM cycles without boundary warnings; overall fit was excellent (−2 LL=11 2020.6, CFI=0.98, RMSEA=0.031, SRMR=0.04), exceeding commonly accepted cut-offs (CFI ≥0.95, RMSEA ≤0.06).48 Discrimination parameters (a) ranged from 0.47 to 2.04 (mean=1.10). Two QoL items were highly informative (a ≥1.70), nine items provided moderate information (0.65≤a <1.70), and the ‘medical errors’ stress item showed low information (a <0.65).66 Thresholds (b₁–b₄) spanned −5.2 to +3.3, ensuring sensitivity across impaired and optimal ranges (online supplemental Table S4). The test information curve (online supplemental Figure S2) exhibited a broad plateau (θ ≥4.3) from θ=−1.5 to +1.8, corresponding to marginal reliability ρ≈0.80–0.83.54 Information remained >2.0 (ρ≈0.67) at θ ≈±3.5, indicating adequate precision even in extreme tails. EAP θ scores for all 2005 respondents were approximately normal (mean 0.04, SD 0.96; skew −0.11; kurtosis −0.14) and were therefore transformed to T-scores (M=50, SD=10) and subsequently to the 0–100 centesimal scale. Weighting did not materially affect item functioning. Average factor loading shifts were ≤0.03 and global fit indices changed by <0.005 (table 4), remaining within accepted thresholds (CFI/TLI ≥0.95; RMSEA ≤0.06; SRMR ≤0.08).48 Table 4Exploratory and confirmatory solutions: unweighted versus post-stratified (weighted)AnalysisMetricUnweightedWeightedΔEFAVariance explained (%)72.171.6−0.5Mean absolute loading shift—0.03—CFA (13 item)CFI0.9610.958−0.003TLI0.9530.950−0.003RMSEA0.0410.042+0.001SRMR0.0460.047+0.001Variance explained refers to the proportion of common variance accounted for by the factors, calculated as the sum of squared standardised loadings divided by the total common variance. Loading shift represents the average absolute difference between unweighted and weighted standardised loadings across the 13 items. Δ=weighted—unweighted. All fit indices remain within recommended limits, confirming that poststratification weights do not compromise latent structure integrity.CFA, confirmatory factor analysis; CFI, Comparative Fit Index; EFA, Exploratory Factor Analyses; RMSEA, Root Mean Square Error of Approximation; SRMR, Standardised Root Mean Square Residual; TLI, Tucker-Lewis Index. Exploratory and confirmatory solutions: unweighted versus post-stratified (weighted) Variance explained refers to the proportion of common variance accounted for by the factors, calculated as the sum of squared standardised loadings divided by the total common variance. Loading shift represents the average absolute difference between unweighted and weighted standardised loadings across the 13 items. Δ=weighted—unweighted. All fit indices remain within recommended limits, confirming that poststratification weights do not compromise latent structure integrity. CFA, confirmatory factor analysis; CFI, Comparative Fit Index; EFA, Exploratory Factor Analyses; RMSEA, Root Mean Square Error of Approximation; SRMR, Standardised Root Mean Square Residual; TLI, Tucker-Lewis Index. Invariance Sequential multigroup CFA supported supported configural, metric, scalar and strict invariance across both sex and graduation year quartiles, with ΔCFI ≤0.002 and ΔRMSEA ≤0.003—satisfying the cut-offs proposed by Chen for invariance retention.55 These results indicate that the three-factor model operates equivalently across groups with respect to factor loadings, intercepts and residual variances, thereby allowing meaningful comparisons of structural relationships between latent variables across sex and graduation cohorts. Nonetheless, full equality of latent means was not confirmed for either stratification. For sex, the ΔCFI was 0.005—within the acceptable threshold—yet the χ² difference was significant (p<0.001), and RMSEA increased from 0.049 to 0.055, suggesting only partial support for latent mean invariance. For graduation year quartiles, invariance of latent means was not supported, with ΔCFI=0.025 and ΔRMSEA=0.028, both exceeding decision thresholds. These findings imply that observed differences in latent mean scores—particularly across graduation cohorts—likely reflect true substantive disparities, potentially attributable to career-stage effects or contextual shifts over time. All fit indices and decisions are reported in online supplemental Table S13. Differential item functioning No items exhibited uniform or non-uniform DIF concerning macro-region or gender, with maximum ΔR² values observed at 0.009 (Item 4) and 0.012 (Item 9), well below the established threshold. In race/ethnicity comparisons, Item 7 (‘My institution recognises my professional achievements’) displayed marginal, non-uniform DIF (ΔR²=0.015); however, its removal minimally affected person scores (<0.10 SD), demonstrating negligible practical impact. Regarding age quartile comparisons, Item 5 (‘I feel capable of handling daily demands’) reached ΔR² = 0.018, again below the defined criterion, and purification resulted in minor score changes (<0.07 SD). Collectively, these results support the equivalence of the scale across macro-region, race/ethnicity, gender and age, validating its applicability for unbiased group comparisons. Afya MedQoL index scores Weighted descriptive statistics for the latent well-being scores (n=2005) are reported in online supplemental Table S9. All scores were derived using the GRM and presented both as EAP and as centesimal transformations for interpretability. The global composite showed a mean θ of 0.01 (SD=0.70), corresponding to a transformed score of 67.2 (SD=15.0). Domain-specific scores were highest for QoL (mean=69.5), followed by Institutional Support (64.1) and Perceived Stress (62.5). The model demonstrated strong reliability (global ρ=0.82; domains ρ=0.78–0.83). Test information exceeded four Fisher units between θ=−1.5 and +1.8, encompassing approximately 90% of the sample. Information-weighted factor scores (θmetric) followed an approximately standard normal distribution (μ≈0, σ≈1) for each domain, whereas the global composite was slightly more concentrated (σ≈0.70). After linear conversion to a 0– 100 centesimal scale, weighted means (SD) were 69.5 (15.4) for F1 QoL, 64.1 (16.2) for F2 Institutional Support, 62.5 (15.8) for F3 Perceived Stress (higher scores indicate greater stress; reverse-coded only for the global index) and 67.2 (15.0) for the global index (online supplemental Table 9). Because the transformation was strictly linear, the rank ordering of scores was nearly perfectly preserved (|r|>0.997) across θ, T and centesimal metrics. Cut points adopted from IRT convention (θ±1.4 SD ≈ centile 10 and 80) classified scores ≤30 as critical, 31–69 as intermediate and ≥70 as high well-being.67 Only 9.6% of physicians scored below the critical threshold on the global index, whereas 20.7% occupied the resilience zone. Marginal reliability of the composite was ρ=0.82, exceeding the 0.70 benchmark for group level comparisons.68 Distributional diagnostics (skew=−0.11; kurtosis=−0.14) supported the use of T-scores and centesimal scaling without further transformation.58 Correlations between θ, T and centesimal metrics were ≥0.997 for every domain, confirming that linear rescaling preserved measurement information. The composite’s marginal reliability remained ρ=0.82—well above the 0.70 criterion for group comparisons.68 Test information stayed ≥4 Fisher units from θ=−1.5 to +1.8, thus covering ~90% of physicians. On the 0–100 scale, critically low well-being (≤30 points; θ < −1.4, ~10th percentile) was observed in 9.6% of respondents, while 20.7% scored ≥70 points (θ > +1.4), reflecting high well-being. Approximately 95% of scores fell between 20 and 80 points, indicating an adequate distribution for both screening and monitoring purposes (figure 1). Stacked ‘thermometer’ heat bars for the Afya MedQoL Index. Stacked continuous ‘thermometer’ density plots showing the distribution of physicians along the 0–100 score continuum for the global composite index (top) and the three domains: Quality of Life (F1), Institutional Support (F2) and Perceived Stress (F3). Colour intensity represents the normalised kernel density estimate for each score (0=no physicians, 1=highest concentration). Convergent validity with WHOQOL-BREF WHOQOL-Bref data were available for 98% of participants (median item missingness=1.1%). Pearson correlations between the AfyaMedQoL Index global score and WHOQOL-Bref domains were strong (|r|≥0.55) and highly significant (p<0.001), meeting the apriori criterion of r ≥0.50 for strong convergence.62 The highest concordance occurred with the Psychological domain (r=0.65) and the lowest with Social Relationships (r=0.55). At factor level, F1 QoL correlated very strongly with Physical (r=0.69) and Psychological (r=0.73) WHOQOL domains (online supplemental Table S8). All coefficients exceeded the prespecified 0.50 threshold, corroborating convergent validity between the Afya MedQoL Index and the legacy WHOQOL-Bref. Concordance remained virtually unchanged when θ or T metrics were analysed (|r|≥0.997), confirming that linear rescaling preserved psychometric information. Sex differences Weighted analyses revealed no statistically significant difference in the global index (Δ=−0.8 points; p=0.16; d=−0.05). However, women reported Perceived Stress (F3) scores 5.6 points higher than men (p<0.001; d=−0.28), a small to medium effect.62 QoL and Institutional Support did not differ materially (table 5). Weighting on sex×region altered no mean difference or effect size by more than 4% (online supplemental Table S5), demonstrating robustness. Table 5Weighted Afya MedQoL Index scores by sexScore (0–100)Male(n=975)Female(n=1028)Δ95% CIP valueCohen’s dGlobal index66.8±15.267.6±14.8−0.8−2.0 to 0.40.162−0.05F1 Quality of Life69.8±15.669.1±15.2+0.7−0.9 to 2.30.364+0.04F2 Institutional Support65.5±16.763.9±15.8+1.6−0.2 to 3.40.078+0.08F3 Perceived Stress59.4±16.665.0±14.9−5.6−7.4 to −3.8<0.001−0.28Δ=mean difference (male–female); positive values favour males. Effect size interpretation: small |d|≥0.20, medium ≥0.500.62 Weighted Afya MedQoL Index scores by sex Δ=mean difference (male–female); positive values favour males. Effect size interpretation: small |d|≥0.20, medium ≥0.500.62 Age, graduation year and income Perceived Stress (F3) declined linearly with age quartile and graduation year cohort (η²=0.09 and 0.07, respectively), whereas the global index showed no age trend. Income increased global scores up to the third quartile (BRL 17 501–27 500) but plateaued thereafter, suggesting a sufficiency threshold around BRL ≈ 25 000 per month. Global well-being was essentially stable across age and income strata, whereas perceived stress declined meaningfully with increasing time since graduation. Specifically, the weighted global score averaged 67.4 points in the youngest age quartile (23–29 years) and 65.7 points in the oldest (46–94 years); the linear trend was not significant (p=0.075) and the effect size was trivial (η²=0.006). By contrast, Perceived Stress fell from 56.8±20.2 in physicians who graduated after 2022 to 40.7±19.9 in those who finished medical school before 2008—a moderate cohort effect (η²=0.070, p<0.001). These findings, presented in online supplemental Table S11, suggest that while age and income do not substantially impact perceived global well-being, career stage—captured by graduation year—has a notable effect on perceived stress, likely reflecting greater work-related pressures and emotional demands among younger physicians. Income quartiles showed virtually no gradient in the global index (means 66.9–66.7 points; p=0.764, η² <0.001), reinforcing that overall well-being plateaus once basic financial sufficiency is reached. Differences in global scores across medical specialties and workplace settings were analysed using the sample mean as the reference point (Internal Medicine and public hospitals, respectively). The results show that physicians in Dermatology, Ophthalmology and Radiology/Pathology reported significantly higher QoL and well-being scores compared with the overall sample, with positive deviations of+8.2, +6.9 and+5.5 points, respectively. In contrast, specialties such as Family Medicine (–4.9), Surgical fields (–3.6) and Obstetrics and Gynaecology (–1.5) had lower scores, with CIs indicating statistically significant differences in some cases. Regarding the workplace setting, professionals working predominantly in private clinics (+6.6) or via telemedicine (+5.4) reported higher levels of well-being than those working in public hospitals, primary care centres (UBS) or private hospitals, whose scores were close to or below the sample average. These findings, detailed in online supplemental Table S6 and S7, suggest that both medical specialty and workplace setting significantly influence physicians’ overall perception of well-being—likely reflecting variations in job demands, professional autonomy and institutional support. Income satiation analysis Generalised additive splines confirmed a plateau in the income–well-being curve at ≈ BRL 25 000 per month for the global index; domain-specific thresholds varied modestly (online supplemental Table S10). Adjusting for weekly hours, household size and specialty shifted plateau by ≤BRL 1000, underscoring the robustness of the saturation point (figure 2). Weighted income well-being curves (monthly income). Hexbin scatter plots (n=2005; design weights) with cubic spline fit (black line, 95% CI shading) for the global Afya MedQoL Index and the three domains. The dashed vertical line marks the inflection at ≈ BRL 25 000 per month, above which the marginal gain is <0.3 points per BRL 1000 (R$=BRL). Multivariable adjustment—introducing weekly workload, household size and specialty into the spline models—left the income thresholds for well-being virtually unchanged. Before adjustment, the global index flattened at BRL 25 000 per month; after covariate control, the plateau shifted downward by only BRL 1000 to BRL 24 000, a 4% reduction. The pattern was identical for Qol, whose saturation point moved from BRL 22 000 to BRL 21 000. For Institutional Support, the threshold rose marginally from BRL 29 000 to 30 000, while Perceived Stress remained unchanged at BRL 15 000. Because the plateau was defined as the first income value at which the spline derivative fell below ±0.003 points per 100 BRL (≈0.3 units), these ±1000 BRL shifts are well within the model’s resolution, confirming that confounder adjustment does not materially alter the income satiation landmarks. Additional covariate effects indicate that every 10 hours/week above the median 48 hours reduced the global score by 1.9 points (p<0.001), while each household member increased it by 0.4 points (p=0.04). Specialty modified only the F2 plateau, surgeons requiring~BRL 1000 more for satiation. Detailed diagnostics are provided in figure 3, which juxtaposes the raw and multivariable adjusted income splines for all domains. Observations are shown as hexagonal density bins in grayscale (darker=denser). The solid line is a cubic regression spline (df=6) with a 95% confidence band (shaded). Vertical dashed lines mark the hypothesised thresholds (BRL 25 000/month in A; 150 BRL/hour in B–E). Hourly income was trimmed to 0–400 BRL/hour to exclude extremes. Monthly income categories were converted to BRL mid-points. All outcomes are scaled 0–100 (higher values indicate better well-being for Global, F1 and F2; higher values indicate greater stress for F3). Across domains, the smoothed associations flatten beyond ~BRL 25 000/month and~150 BRL/hour, a pattern consistent with income satiation. For F3, the near-flat curve beyond the threshold indicates minimal additional reductions in perceived stress at higher earnings. Domain-specific patterns were also observed: QoL rises steeply up to ≈ BRL 22 000 and flattens earliest; Institutional Support remains almost horizontal, indicating minimal income responsiveness; Perceived stress falls sharply until ≈ BRL 15 000 and then levels off, suggesting stress relief only among lower earners; while the global Afya MedQoL Index combines these tendencies, increasing modestly to ≈ BRL 25 000 and gaining <0.3 points per BRL 1000 thereafter. The mirror image similarity between monthly and hourly columns rules out artefacts related to income denomination, and the overlap of 95% confidence ribbons indicates statistical equivalence except in the lowest income decile, where workload adjustment slightly steepens the ascent. Income and well-being among Brazilian physicians—evidence of income satiation. (A) Global well-being versus net monthly income (BRL), (B) Global well-being versus net hourly income (BRL/hour), (C) Quality of Life (F1) versus net hourly income (BRL/hour), (D) Institutional Support (F2) versus net hourly income (BRL/hour) and (E) Perceived Stress (F3) versus net hourly income (BRL/hour). BRL, Brazilian real; F1, Quality of Life; F2, Institutional Support; F3, Perceived Stress; h, hour. Weighted generalised additive models explained 30%–32% of variance in the global Afya MedQoL Index as well as in QoL and Perceived Stress, but only 8% in Institutional Support. Below BRL 25 000 per month, each additional BRL 1000 was associated with a significant rise of 0.68 points in the global score (95% CI 0.05 to 1.31) and 0.71 points in QoL, while Perceived Stress fell by 0.76 points (−1.33 to −0.19); the slope for Institutional Support was trivial (+0.22; CI spans zero). Above this income, all slopes converged on zero and the first derivative test identified plateaux at BRL 24 900 (22 400–27 600), for the global Afya MedQoL Index BRL 25 100 (22 500–27 800) for QoL and BRL 24 700 (21 900–27 500) for Perceived Stress; no measurable saturation emerged for Institutional Support. Truncating extreme survey weights (>4) left adjusted R2 values unchanged and shifted plateau estimates by no more than±BRL 200, confirming that the findings were not driven by high-weighted cells or residual imbalance. Collectively, these results demonstrate that the weighted dataset is demographically representative, that physician well-being increases with income only up to the upper middle class benchmark of~BRL 25 000 per month, and that this ceiling effect remains stable under rigorous sensitivity checks. Discussion This study developed the first perception-based QoL scale specifically tailored to physicians, integrating exploratory and CFAs with multidimensional graded response calibration.52 The final 13-item instrument captures three correlated domains—QoL (F1), Institutional Support (F2) and Perceived Stress (F3)—and yields a global composite weighted by Fisher information, satisfying contemporary recommendations for brevity, precision and adaptivity.8 67 After post-stratification to the 2025 Brazilian medical demographic census margins,7 the weighted sample replicated the national distribution of physicians by sex (51% women), age ≥55 years (26%) and macroregion (North 6%, Northeast 24%, Centre West 12%, Southeast 46%, South 12%). Standardised differences (Cohen’s h) for each calibration variable decreased from ≤0.26 to ≤0.02, well below the 0.20 threshold for negligible imbalance.69 Domain means shifted by ≤0.5 points and effect sizes by ≤4%, confirming external validity. Three overarching patterns emerged. First, perceived stress dominated variance, age gradients and gender differences. Women scored 5.6 points higher on F3, a small to medium effect that aligns with meta-analytic data on burnout prevalence among female physicians.1 6 Structural factors such as unequal domestic workload and heightened emotional labour are likely to explain this gap.17 70 Stress also declined almost linearly with the graduation cohort, echoing longitudinal evidence that psychological distress peaks early in medical careers.71 Second, Institutional Support exhibited the lowest mean and widest dispersion. The most discriminative items involved non-punitive responses to errors and access to formal mental health support pillars of a ‘just culture’.7274 Private outpatient clinics and telemedicine practices scored 5–7 points higher than public hospitals, consistent with data linking autonomy and flexible scheduling to reduced emotional exhaustion.75 Third, the marginal utility of income showed a clear upper limit. Weighted additive models explained 30%–32% of variance in the global index and domains F1 and F3, but only 8% in F2. Below BRL 25 000 per month, income was positively associated with global QoL (+0.68 points per BRL 1000; 95% CI 0.05 to 1.31) and negatively associated with stress (–0.76; –1.33 to –0.19). Above this threshold, all slopes converged on zero, producing a plateau at BRL 24 900 (global), BRL 25 100 (QoL) and BRL 24 700 (Perceived Stress). Truncating extreme survey weights (>4) altered these estimates by at most±BRL 200, attesting to robustness. This threshold aligns with international evidence suggesting that subjective well-being plateaus around US$ 60 000–75 000 PPP in upper-middle-income economies76 and supports the hypothesis that, once material comfort is secured, time control and work context eclipse monetary reward.77 Several divergences persisted after weighting. Women showed a 5.6-point excess in Perceived Stress, indicating a genuine gender gap. Long work weeks (≥60 hours/week) were associated with an 8.8-point higher F3 score (greater perceived stress). This pattern is dialogues with national census evidence that approximately 31% of physicians report working ≥60 hours/week.7 Ambulatory specialties (eg, dermatology) were overrepresented and consistently showed higher well-being scores and retained higher well-being scores, underscoring true specialty effects. This study contributes to the growing body of evidence on physician well-being by delivering a rigorously validated, domain-specific instrument—the Afya MedQoL—and by situating Brazilian findings within international trends and the 2025 National Medical Demography Census. Three broad themes emerge: (1) The centrality of perceived stress, (2) The pivotal role of institutional context and (3) The diminishing marginal returns of income on QoL. Perceived Stress showed the largest variance, the steepest age gradient and the most consistent sensitivity to occupational exposures. The 6-point sex gap in Perceived Stress replicates meta-analytic evidence that female physicians have 20% to 60% higher odds of burnout and depressive symptoms, even after adjusting for specialty and workload.1 2 4 Two complementary hypotheses may help explain this discrepancy. First, Brazilian micro-data confirm that women still undertake 1.9 hours more unpaid domestic work per day than men,78 amplifying total role overload. Second, qualitative studies describe higher exposure of female doctors to patient-related emotional labour and gender-biased micro-aggressions, both recognised precursors of exhaustion.70 79 Perceived stress scores declined nearly linearly with age and years since graduation, echoing longitudinal observations from Canadian residents where psychological distress peaks in postgraduate year 1 and normalises only after year 5.71 Lower clinical autonomy, higher evaluative pressure and limited coping repertoires in early career stages provide a plausible mechanism within the demand–control framework.80 Brazilian residency programmes average 60 to 80 hours/week with half-day post-call rest still commonplace,81 making long working hours an entrenched norm rather than an individual choice. Institutional Support not only registered the lowest mean but also the broadest dispersion, signalling that organisational climate is highly heterogeneous across practice settings. Items with the greatest discrimination—non-punitive handling of adverse events and formal mental-health assistance—mirror the twin pillars of ‘just culture’ and ‘second-victim’ support widely advocated in patient-safety literature.73 74 Private outpatient clinics and predominantly telemedicine practices scored 6 points and 5 points higher, respectively, on the global index; both environments afford greater control over workflow and reduce exposure to rigid hierarchies—conditions linked to lower emotional exhaustion.75 Conversely, public hospitals, where bureaucratic constraints and punitive cultures remain prevalent, were positioned at the lower end of the distribution. The 9-point increase in stress observed beyond 60 hours per week aligns with a non-linear dose-response pattern, in which the risk of burnout doubles beyond 60 hours, triples beyond 74 and quadruples beyond 84 hours, as demonstrated in a large cross-sectional study of 2081 healthcare workers in Taiwan.82 Adjusted analyses confirm a robust income satiation threshold of roughly BRL 25 000 per month (≈ US$ 5000) for Brazilian physicians, in line with cross-national evidence that subjective well-being plateaus at the equivalent of USD 60 000–75 000 purchasing power dollars in upper-middle-income economies.76 The persistence of this plateau after controlling for weekly work-hours, household size and specialty underscores the view that, once basic material comfort is secured, time and workplace context eclipse further monetary gains as determinants of professional QoL.77 Each additional ten clinical hours per week reduced Overall QoL by almost two points—an effect similar to the reported in US physicians.2 Perceived Stress flattened at the lowest income level (~BRL 15 000 per month), replicating findings that financial insecurity is one of the first stressors to abate once comfort thresholds are crossed.83 Beyond that threshold, non-financial interventions such as mindfulness training, peer coaching and schedule autonomy—deliver substantially larger stress reductions than any conceivable salary increment. Because post-plateau income produced <0.02 QoL points per additional BRL 1000, across-the-board salary increases above approximately BRL 25 000 per month are inefficient substitutes for workload control and institutional investment. Notably, a 10 hours/week workload reduction (≈ –1.9 QoL points gained) ‘costs’ the system nothing in direct salary yet outperforms a BRL 10 000 raise (+0.9 QoL points). This aligns with evidence that organisational level changes—reducing clerical burden, improving staffing ratios—improve physician well-being.84 85 Health-system planners might therefore channel limited resources towards structural and psychosocial interventions rather than blanket pay rises, at least for physicians already earning above the identified thresholds. Ambulatory specialties—dermatology, ophthalmology, radiology/pathology—led the well-being rankings, corroborating surveys linking controllable lifestyles to higher life satisfaction.2 Self-reported psychiatric diagnoses and high recent negative affect produced the largest effect sizes (d≈0.6–0.8), dwarfing demographic predictors—a pattern repeatedly documented in systematic reviews.8688 The reciprocal relationship between mood disorders and burnout suggests that reactive interventions targeting stress alone may be insufficient; integrated models that encompass early detection, stigma reduction and formal treatment pathways might be recommended.1 The advantage associated with predominantly remote practice supports preliminary data that telemedicine reduces time pressure and work–family conflict, provided adequate technical support is available.89 90 Hypothetically, asynchronous communication and patient pre-triage may buffer cognitive load and allow physicians to sequence tasks according to personal chronotypes, an effect worthy of experimental confirmation.91 Beyond substantive findings, this work advances measurement science in three respects. First, it offers a multidimensional, culturally adapted tool validated on a nationally stratified physician sample—something only a handful of instruments achieve.9 13 Second, the graded response calibration allows short-form creation and computerised adaptive testing without loss of comparability, responding to critiques that existing tools are too lengthy for routine monitoring.8 Third, the weighting of the global score by item discrimination aligns the composite with regions of maximal measurement precision, an approach rarely operationalised outside research settings.67 Policy implications emerge from these patterns. (1) Because perceived stress was most sensitive to sex, age and workload, psychosocial interventions should prioritise female and early-career physicians, (2) Well-being plateaued above BRL 25 000/month, whereas workload beyond 60 hours/week incurred a ∼9-point stress penalty, supporting calls for work-hour caps, (3) Telemedicine and private outpatient practice each increased global well-being by >5 points, aligning with census projections that flexible practice models will expand and (4) Self-reported mental disorders and high negative affect produced the largest effect sizes (d≈0.6–0.8), highlighting the urgency of nationwide mental-health programmes. Limitations This study should be interpreted in light of its limitations. The cross-sectional design precludes causal inference; however, the magnitude of effects and their consistency with longitudinal data strengthen the plausibility of the observed associations. The survey was disseminated online, raising concerns about self-selection bias; however, post-stratification to census margins altered domain means by no more than 0.5 points, suggesting minimal impact. Positive psychology constructs such as flow, purpose and workplace spirituality—shown to buffer burnout10—were not included and merit integration in future item bank expansions. Conclusion The Afya MedQoL Index represents the first Brazilian instrument grounded in IRT and calibrated with Samejima’s GRM to assess physician well-being across three psychometrically distinct domains: dQoL, Institutional Support and Perceived Stress. In a nationally weighted cohort of 2005 practitioners, the scale demonstrated a robust three-factor structure, marginal reliabilities ≥0.80 within the core latent interval (θ≈–1.5 to +1.8) and external validity after alignment to 2025 national medical census margins, with domain means shifting by <0.5 points and effect sizes by <4%—well below thresholds for practical importance. Three key findings emphasise. First, Perceived Stress emerged as the most responsive domain showing a small-to-moderate disadvantage for women, early-career physicians and those working 60 or more hours per week, echoing global burnout literature that cites role overload and gendered microaggressions as key drivers.1 4 79 82 86 87 Second, Institutional Support exhibited wide heterogeneity: private outpatient and telemedicine settings outperformed public hospitals by 5–7 points, consistent with evidence that autonomy and a ‘just culture’ buffer emotional exhaustion.7274 Third, global well-being plateaued at ≈ BRL 25 000 per month, aligning with the income satiation hypothesis observed in upper-middle-income economies and underscoring the supremacy of non-pecuniary factors—autonomy, purpose and social connectedness—once material sufficiency is achieved.77 These findings support the following policy recommendations: Prioritise stress reduction programmes for women and physicians in early career stages. Prioritise stress reduction programmes for women and physicians in early career stages. Cap weekly clinical hours at ~60, in line with international guidance. Cap weekly clinical hours at ~60, in line with international guidance. Promote non-punitive, supportive organisational cultures and offer formal ’second-victim’ support services and Promote non-punitive, supportive organisational cultures and offer formal ’second-victim’ support services and Integrate systematic mental health screening and treatment into occupational well-being initiatives. Integrate systematic mental health screening and treatment into occupational well-being initiatives. Future work should evaluate Afya MedQoL Index responsiveness in longitudinal designs, validate translations in other languages and triangulate self-report data with objective indicators such as electronic health record activity or physiological stress markers. Supplementary material