Patient Perceptions Across a Fixed Sequence of Physician Attire and Surgical-Context Images in Urology: An Image-Based Cross-Sectional Survey

Abstract

Background

Physician attire is associated with patient trust and confidence, yet it has rarely been examined in urology or in the surgical context in which the physician is seen.

Aims

To examine how physician attire and the displayed surgical context are associated with patients’ initial perceptions of urologists using standardized, artificial intelligence (AI)-generated images.

Study Design

Cross-sectional study.

Methods

A total of 220 adult urological patients rated 10 standardized AI-generated physician images, presented in the same fixed order, using 11 5-point Likert items and one binary acceptance item. Five images varied by attire (casual, business suit, suit with a white coat, scrubs, and scrubs with a white coat), and five varied by surgical context (neutral, open, endourological, laparoscopic, and robotic). Images were matched to each participant’s stated physician-gender preference; the male-image stream (n = 189) formed the primary analysis, and the female-image stream (n = 31) formed a secondary analysis. Within-subject comparisons used the Friedman test with Bonferroni adjustment across outcomes, Page’s trend test for the ordered attire hypothesis, and Wilcoxon signed-rank tests. Treatment acceptance was modeled using generalized estimating equations (GEEs) with an exchangeable working correlation.

Results

Ratings varied across the fixed, ordered attire sequence for every perception item (Friedman tests, all Bonferroni-adjusted p < 0.001); Page’s trend tests for the two primary outcomes were also significant (both Bonferroni-adjusted p < 0.001). Because the attire category was perfectly confounded with presentation position, this pattern cannot be attributed to attire independently of order. For male images, mean trust increased from 2.98 ± 1.34 (casual clothing) to 4.26 ± 0.81 (scrubs with a white coat), and choice likelihood increased from 3.07 ± 1.37 to 4.29 ± 0.84; observed binary treatment acceptance increased from 56.1% to 97.4%. In the adjusted GEE model, acceptance was higher at position 5 (scrubs with a white coat) than at position 1 (casual clothing) [odds ratio, 39.1; 95% confidence interval (CI), 13.1-116.7; adjusted risk difference, +40.3 percentage points; 95% CI, 32.6-48.0]. By contrast, only leadership and attractiveness remained significant among the surgical context comparisons after adjustment, and all context effect sizes were small (Kendall’s W ≤ 0.03). Because the context images also differed in several unintended visual respects beyond the operative setting, this comparison is restricted to the specific images shown.

Conclusion

In this standardized, image-based survey, ratings varied systematically across the fixed, ordered image sequence, with the largest and most consistent differences appearing across the images intended to vary in physician attire. Because the attire category was perfectly confounded with presentation position, this association cannot be attributed to attire alone. Associations with the displayed surgical context were small under the conditions tested and are further limited by unintended differences among those images beyond the operative setting.

INTRODUCTION

The patient-physician relationship is a central determinant of patient experience, satisfaction, and treatment adherence.¹ Much of this relationship is established during the first moments of contact: people form trait judgments, such as trustworthiness, from a face within a fraction of a second.² Among the non-verbal cues available at first contact, physician attire is one of the most salient, communicating competence and professionalism,³ and clothing carries symbolic weight that can shape behavior and self-perception.⁴

Attire shapes how patients perceive their physicians. In a classic photograph-based survey, professional attire with a white coat elicited the greatest trust and confidence,⁵ and comparable patterns have been described across diverse healthcare settings and countries.^{6, 7} At the same time, surgical scrubs have become increasingly accepted and, in some procedural and inpatient contexts, are actively preferred.⁸ This varies internationally: scrubs were the most preferred attire among Italian and Swiss patients in a 4-country comparison, whereas formal attire was preferred in Japan and the USA.⁹

Most prior work shares two limitations. First, attire has chiefly been studied in non-procedural specialties such as pediatrics and obstetrics and gynecology,^{10, 11} with comparatively little attention to surgical disciplines.¹² Second, the surgical or technological context in which a physician is encountered has rarely been examined as a perceptual factor in its own right. Urology is a highly procedural, technology-driven specialty in which patients may encounter their surgeon alongside open, endoscopic, laparoscopic, or robotic platforms. Public attitudes toward such platforms are themselves variable and are frequently shaped by incomplete understanding.^{13, 14}

Photographs of different physicians introduce confounding by face, age, expression, and background.¹⁵ Artificial intelligence (AI)-generated images reduce face-related variation while permitting more controlled manipulation of attire and context, although incidental features were not fully standardized across the surgical context images. We therefore examined how attire and surgical context are associated with patient trust, treatment choice, and related perceptions in urology.

MATERIALS AND METHODS

Study design and participants

Data were collected between October 1, 2025, and February 28, 2026. Adult (≥ 18 years) native Italian-speaking patients attending urology outpatient clinics and inpatient wards were invited consecutively (convenience sampling) at the Department of Urology, ASST Lariana, Sant’Anna Hospital, Como (Lombardia, northern Italy). Each participant completed a questionnaire hosted on Google Forms (Google LLC, Mountain View, CA, USA), which was accessed by scanning a QR code on their smartphone. Participants were approached individually by a study team member who confirmed single, in-person completion; only fully completed, non-duplicate questionnaires were included in the dataset. Participation was voluntary and anonymous, and electronic informed consent was obtained beforehand. Individuals unable to provide consent or complete the questionnaire were not eligible. No a priori sample size or power calculation was performed; the sample reflects patients recruited during the predefined study period, and the number who declined was not recorded. Participant flow is summarized in Figure 1. The study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology guidance for cross-sectional studies (Supplementary Checklist).

Figure 1.

Flow of participants through the study.

Development of physician images

Ten standardized physician images were created for each physician-gender in August 2025 using a generative AI image model (Gemini 2.5 Flash Image, Google DeepMind, Mountain View, CA, USA; model identifier, gemini-2.5-flash-image-preview), which preserves subject identity across successive edits. A single base identity was generated for each gender and held constant across all variants, with a neutral expression, similar apparent age (30-40 years), minimal grooming, and no jewelry, brands, tattoos, or logos. Five attire images (Table 1) were produced against an identical neutral studio background, with the physician standing and facing the camera directly. Five further images were intended to show the physician in fixed surgical attire across five operative context scenes (Table 2), but post-hoc inspection identified incidental differences beyond the intended manipulation—in attire, pose, and framing—detailed in the Supplementary Methods. These included the robotic-console scene, in which the physician was shown seated and in profile rather than standing and facing the camera. Comparisons among the context images therefore reflect these specific composite scenes rather than an isolated manipulation of the operative setting (Discussion). No patient was shown. Images were generated at 1024 × 1024 pixels using a fixed random seed for each identity and a locked prompt library. Before use, all images were independently reviewed by 3 research team members against predefined criteria (correct attire, recognizable setting, neutral affect, and absence of extraneous or branding cues), and images failing any criterion were regenerated. This review was an internal check rather than a formal manipulation check with independent raters, which is a limitation. Because attire was varied within a neutral setting and context was varied within a nominally fixed attire, the design was not a full attire × context factorial. Full generation specifications and the complete image set are provided in the Supplementary Material (Supplementary Figures 1 and 2, Supplementary Methods).

TABLE 1. Ratings Across the Five Images in the Fixed Attire Sequence, Male-Image Stream (n = 189). Values are Mean ± Standard Deviation on a 1-5 Likert Scale.

Perception item	Casual	Suit	Suit + coat	Scrubs	Scrubs + coat	χ 2	p*	W	Page z
Trust*	2.98 ± 1.34	3.32 ± 1.20	3.89 ± 0.96	4.04 ± 0.89	4.26 ± 0.81	309.5	< 0.001	0.409	13.5
Choice likelihood*	3.07 ± 1.37	3.44 ± 1.25	3.89 ± 0.98	4.08 ± 0.94	4.29 ± 0.84	272.5	< 0.001	0.360	12.5
Technical competence	3.06 ± 1.34	3.35 ± 1.24	3.89 ± 1.01	4.04 ± 0.92	4.24 ± 0.83	250.6	< 0.001	0.332	12.1
Professionalism	2.97 ± 1.40	3.45 ± 1.29	3.97 ± 0.98	4.06 ± 0.92	4.31 ± 0.83	258.8	< 0.001	0.342	12.4
Communication comfort	3.05 ± 1.36	3.32 ± 1.21	3.92 ± 0.98	4.01 ± 0.95	4.27 ± 0.87	240.8	< 0.001	0.319	12.1
Cleanliness/hygiene	3.14 ± 1.41	3.42 ± 1.25	3.98 ± 1.00	4.02 ± 0.94	4.31 ± 0.88	205.2	< 0.001	0.271	11.0
Leadership	2.87 ± 1.31	3.31 ± 1.28	3.83 ± 0.95	3.91 ± 0.98	4.15 ± 0.87	244.7	< 0.001	0.324	12.0
Experience	2.92 ± 1.31	3.29 ± 1.24	3.87 ± 0.98	3.98 ± 0.94	4.19 ± 0.88	278.5	< 0.001	0.368	12.6
Uneventful surgery	2.94 ± 1.31	3.31 ± 1.26	3.81 ± 0.95	3.93 ± 0.96	4.17 ± 0.91	234.3	< 0.001	0.310	11.9
Comfort when treated	2.98 ± 1.35	3.36 ± 1.22	3.94 ± 0.98	3.99 ± 0.96	4.20 ± 0.85	242.4	< 0.001	0.321	11.6
Attractiveness	2.67 ± 1.34	3.08 ± 1.34	3.48 ± 1.18	3.60 ± 1.18	3.78 ± 1.18	226.1	< 0.001	0.299	11.3

TABLE 2. Ratings Across the Five Images in the Fixed Context Sequence, Male-Image Stream (n = 189). Values are Mean ± Standard Deviation on a 1–5 Likert Scale.

Perception item	Neutral	Open	Endo	Lap	Robotic	χ 2	p*	W
Trust*	4.03 ± 1.00	4.07 ± 1.09	4.00 ± 1.07	3.94 ± 1.23	4.11 ± 1.07	10.7	0.329	0.014
Choice likelihood*	4.04 ± 1.05	4.08 ± 1.12	4.05 ± 1.09	3.93 ± 1.20	4.08 ± 1.11	8.9	0.708	0.012
Technical competence	4.05 ± 1.05	4.10 ± 1.11	4.02 ± 1.09	3.90 ± 1.19	4.11 ± 1.10	11.1	0.278	0.015
Professionalism	4.08 ± 1.04	4.04 ± 1.14	4.03 ± 1.12	3.95 ± 1.23	4.11 ± 1.09	1.9	1.000	0.003
Communication comfort	4.03 ± 1.03	4.04 ± 1.11	3.94 ± 1.10	3.84 ± 1.24	4.02 ± 1.11	9.6	0.527	0.013
Cleanliness/hygiene	4.05 ± 1.04	4.05 ± 1.12	4.01 ± 1.13	3.96 ± 1.18	4.12 ± 1.10	7.5	1.000	0.010
Leadership	3.91 ± 1.04	4.04 ± 1.13	3.95 ± 1.10	3.92 ± 1.20	4.08 ± 1.13	18.4	0.012	0.024
Experience	3.96 ± 1.06	4.04 ± 1.11	3.98 ± 1.10	3.89 ± 1.21	4.04 ± 1.13	11.3	0.255	0.015
Uneventful surgery	3.97 ± 1.05	4.02 ± 1.13	3.98 ± 1.11	3.86 ± 1.21	4.03 ± 1.13	10.9	0.304	0.014
Comfort when treated	4.02 ± 1.00	3.99 ± 1.14	3.96 ± 1.11	3.94 ± 1.19	4.08 ± 1.10	5.2	1.000	0.007
Attractiveness	3.50 ± 1.32	3.67 ± 1.29	3.65 ± 1.29	3.59 ± 1.35	3.68 ± 1.35	22.0	0.002	0.029

Survey instrument and outcomes

The questionnaire was developed by the research team and administered in Italian (Supplementary Questionnaire). The 11 perception items were generated by the authors from domains identified in the prior attire literature; no formal psychometric validation (content-validity indexing, construct validation, or factor analysis) was undertaken, so the instrument is purpose-built rather than validated. It was translated into Italian and independently back-translated; piloting with 15 respondents confirmed comprehensibility without subsequent changes, and these responses were retained in the main dataset. Participants first reported age, sex, country of residence, education, occupation, perceived household income, self-rated health literacy (a single 1-5 item and not a validated instrument), history of urological surgery, and preferred urologist gender. For each image, they rated 11 statements on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree), listed in Table 1 (choice likelihood denotes willingness to choose the physician for a urological procedure). They then answered a separate binary question on whether they would proceed with treatment by the depicted physician (yes/no); throughout this report, “choice likelihood” denotes the Likert item, and “treatment acceptance” denotes the binary item. Trust and choice likelihood were prespecified as primary outcomes; attractiveness was treated as a distinct, non-professional construct, excluded from the composite score below, and reported separately. All participants viewed the images in the same fixed order—the five attire images first, from casual to scrubs with a white coat, followed by the five context images, from the neutral view to the robotic console—rating each image on every item before the next image appeared. The order was identical in both streams.

Image assignment by physician-gender preference

Images were presented in a manner matched to each participant’s stated physician-gender preference. Participants who preferred a female urologist evaluated the 10 female images, whereas those who preferred a male urologist or expressed no preference evaluated the 10 male images. Image gender was therefore determined by participant preference rather than randomized, and the two streams differed systematically in composition and size. For this reason, the male-image stream was treated throughout as the primary analysis; the female-image stream was reported as a separate, secondary description of ratings within that stream, and no comparison, formal or informal, between the two streams was made or intended. Detailed female-image results are presented in the Supplementary Material to avoid implying such a comparison.

Statistical analysis

Continuous variables are summarized as mean ± standard deviation (SD) and, for perception items, additionally as median (interquartile range) in the Supplementary Material; categorical variables are summarized as frequencies and percentages. Attire and surgical context (five levels each) were compared as within-subject factors using the Friedman test, with Kendall’s W as the within-test effect size; omnibus p values were Bonferroni-adjusted across the 11 outcomes within each arm. Because the Friedman test evaluates any difference rather than an ordered alternative, the ordered hypothesis concerned the 2 prespecified primary outcomes (trust and choice likelihood). Page’s trend test (L) was subsequently used to evaluate this ordered alternative directly and was reported as z, with Bonferroni-adjusted p values for the 2 tests. Corresponding Page tests for the 9 secondary outcomes, the exploratory composite, the secondary female-image stream, and the context-sequence sensitivity check were unadjusted and are reported in the Supplementary Material (Supplementary Table 1) as descriptive rather than confirmatory. Pairwise post-hoc testing was restricted to the 2 primary outcomes and the exploratory composite (Wilcoxon signed-rank test, Bonferroni-adjusted, matched-pairs rank-biserial r). Cronbach’s alpha (α) for the 10 perception items (excluding attractiveness) was computed within each image condition to avoid between-condition inflation, and the interitem correlation matrix is reported. A Global Perception Score, defined post-hoc as an exploratory composite, was calculated as the mean of these 10 items. Treatment acceptance for the attire images was modeled using generalized estimating equations (GEEs) logistic regression with an exchangeable working correlation (preferred to independence given within-participant correlation), with image condition within the fixed attire sequence (reference: position 1, casual clothing), age, participant sex, education, income, and prior surgery as covariates. Results are reported as odds ratios [(ORs), 95% confidence interval (CIs)] and, given near-100% acceptance in the highest categories, as adjusted probabilities and risk differences, the latter with a 95% CI from cluster (participant-level) bootstrap resampling (756/800 successful replicates). Sensitivity analyses refitted the model using an independence working correlation, a random-intercept model (adaptive Gauss-Hermite quadrature), and Firth penalized likelihood fitted without clustering adjustment, with Wald CIs obtained from the inverse penalized information matrix. The Firth model does not account for repeated measurements and serves only as a limited sensitivity check for sparse-data bias, rather than as a clustering-adjusted alternative to the GEE. The 2 streams were analyzed separately and not compared. A p value < 0.05 was considered significant. Analyses were performed using Jamovi (version 2.4; The Jamovi project, Sydney, Australia) and Python 3.11 (Python Software Foundation, Wilmington, DE, USA; custom NumPy/SciPy implementations of Firth regression and the bootstrap; statsmodels for the GEE).

Ethics

According to national regulations governing anonymous questionnaire-based studies that collect no identifiable personal data, formal ethics committee approval was not required for this study. The study was nevertheless conducted in accordance with the Declaration of Helsinki; electronic informed consent was obtained from all participants before initiation, and participants could withdraw at any time without consequence.

RESULTS

Participant characteristics

A total of 220 patients completed the survey (Figure 1; Table 3). The mean age was 52.1 ± 17.5 years (range, 18-91 years). Of these, 189 preferred a male urologist or expressed no preference and evaluated male images (primary analysis), whereas 31 preferred a female urologist and evaluated female images (secondary analysis), yielding 2,200 image ratings.

TABLE 3. Characteristics of the Study Participants (n = 220).

Characteristic	Category	n (%)
Age, years (mean ± SD; range)	-	52.1 ± 17.5 (18-91)
Sex	Male	152 (69.1)
-	Female	62 (28.2)
-	Prefer not to say	4 (1.8)
-	Other	2 (0.9)
Education	Primary	36 (16.4)
-	Secondary	86 (39.1)
-	University	70 (31.8)
-	Postgraduate	28 (12.7)
Residence	Urban	155 (70.5)
-	Rural	65 (29.5)
Perceived household income	Low	52 (23.6)
-	Medium	134 (60.9)
-	High	34 (15.5)
Health literacy (1-5, mean ± SD)	-	3.60 ± 1.15
Prior urological surgery	Yes	100 (45.5)
-	No	120 (54.5)

Ratings across the attire sequence

Ratings differed across the 5 images in the fixed attire sequence for every perception item in the male-image stream (Friedman test, all Bonferroni-adjusted p < 0.001; Table 1). For the 2 primary outcomes, Page’s trend tests supported an increasing association across the sequence (trust, z = 13.47; choice likelihood, z = 12.51; both Bonferroni-adjusted p < 0.001); secondary outcomes showed the same direction descriptively (Supplementary Table 1). Mean trust increased from 2.98 ± 1.34 (casual clothing) to 3.32 ± 1.20 (suit), 3.89 ± 0.96 (suit with coat), 4.04 ± 0.89 (scrubs), and 4.26 ± 0.81 (scrubs with coat); choice likelihood increased in parallel from 3.07 ± 1.37 to 4.29 ± 0.84. Effect sizes for the primary outcomes were moderate (Kendall’s W = 0.41 and 0.36, respectively). Across the pairwise analyses of the 2 primary outcomes and the exploratory composite, all contrasts were significant except the comparison of suit with a white coat versus scrubs for trust (adjusted p = 0.121, r = 0.32). Within the fixed sequence, the white-coat images (positions 3 and 5) received higher ratings than the corresponding earlier suit and scrub images (both p < 0.001, r = 0.88 and 0.67, respectively). Medians and full post-hoc results are provided in Supplementary Tables 2 and 3; this pattern is shown in Figures 2 and 3. Because the presentation order was fixed and identical to the attire sequence, this pattern is reported as an association across the ordered image sequence shown, rather than as an isolated effect of attire.

Figure 2.

Mean ratings (± standard error) of (a) trust and (b) choice likelihood across the five images in the fixed attire sequence, male-image stream (n = 189). Ratings increased across the ordered sequence (Page’s trend test, both primary outcomes Bonferroni-adjusted p < 0.001). Because image condition and presentation position were confounded, this pattern should not be interpreted as an isolated attire effect. Corresponding values for the secondary female-image stream are shown in Supplementary Figure 3 and are not intended for comparison with this figure.

Figure 3.

Radar plot of five perception domains (trust, technical competence, professionalism, leadership, and comfort when treated) across the fixed attire-image sequence, male-image stream. Mean ratings (1-5) expand outward from the casual-clothing image to the scrubs-with-white-coat image. Because the perception items were highly inter-correlated, the domains shown are not independent.

Internal consistency of the 10 perception items was very high when computed separately within each image condition (Cronbach’s α, 0.972-0.989; median, 0.984), and the mean interitem Spearman correlation was 0.83 (range, 0.78-0.91; Supplementary Table 4). These values indicate substantial redundancy consistent with halo-type responding and do not establish a validated unidimensional construct; item-level results should not be interpreted as independent confirmations. For descriptive purposes only, an exploratory Global Perception Score, defined post-hoc as the mean of the 10 items, reproduced the same pattern across the sequence, increasing from 3.00 ± 1.27 for casual clothing to 4.24 ± 0.75 for scrubs with a white coat (Friedman χ² = 357.0, adjusted p < 0.001, W = 0.47; Page z = 16.5, p < 0.001).

Ratings across the surgical context images

Differences across the displayed surgical context were small (Table 2). Mean ratings were uniformly high across the neutral, open, endourological, laparoscopic, and robotic settings, and all effect sizes were negligible (Kendall’s W ≤ 0.03). After Bonferroni adjustment, only leadership (p = 0.012) and attractiveness (p = 0.002) remained significant; trust (p = 0.329) and choice likelihood (p = 0.708) did not. Two design features temper this near-null result. First, the context images were nominally presented in similar surgical attire, and ratings were concentrated within a narrow, high range: the range of condition means for trust was 0.17 points across contexts versus 1.28 across attire, the mean within-participant SD was 0.45 versus 0.71, and 42.8% versus 30.2% of ratings were at the scale maximum. This compression reduces the variance-based Kendall’s W independently of any true difference. Second, the 2 arms were not factorially designed and are described side by side rather than compared statistically. Because the context images also differ in pose, framing, and other incidental features beyond the operative setting, this near-null result reflects these specific scenes and should not be interpreted as evidence that the operative setting itself has no bearing on perceptions.

Findings in the female-image subgroup

Among the 31 participants who evaluated female images, ratings differed across the 5 images in the fixed attire sequence for every perception item (all Bonferroni-adjusted p < 0.001), and descriptive Page tests showed an increasing pattern (all unadjusted p < 0.001). Mean trust increased from 2.61 ± 1.28 for casual clothing to 4.00 ± 1.10 for scrubs with a white coat. No surgical-context comparison reached significance (all adjusted p = 1.000). Because image gender followed participant preference rather than randomization and this stream is secondary and small, these findings are not interpreted confirmatorily and are not compared with the male-image stream; full details are provided in Supplementary Tables 5 and 6, Supplementary Figure 3.

Binary treatment acceptance

Willingness to proceed with treatment, the separate binary outcome, followed the same pattern as the Likert items (Table 4; Figure 4). In the male-image stream, acceptance increased across the 5 images in the fixed attire sequence, from 56.1% (casual clothing) to 69.8% (suit), 93.7% (suit with coat), 95.2% (scrubs), and 97.4% (scrubs with coat) (Cochran Q = 205.8, p < 0.001). Acceptance did not differ across the 5 specific context images (87.3-92.1%; Cochran Q = 5.4, p = 0.246). In the secondary female-image stream, the corresponding attire values were 51.6% to 93.5% (Cochran Q = 31.3, p < 0.001), with no difference across contexts (p = 0.127).

TABLE 4. Willingness to Proceed with Treatment (Binary Outcome), Male-Image Stream (n = 189).

Fixed attire sequence	Casual	Suit	Suit + coat	Scrubs	Scrubs + coat
Would proceed, % yes	56.1	69.8	93.7	95.2	97.4

Figure 4.

Willingness to proceed with treatment (%) across the five images in the fixed attire sequence, male-image stream: observed proportions and probabilities predicted by the adjusted model of Table 5.

Multivariable analysis of treatment acceptance

In the multivariable GEE model restricted to the male-image stream (915 observations from 183 participants; exchangeable working correlation, 0.164), treatment acceptance remained associated with image condition within the fixed attire sequence after adjustment (Table 5). Six participants who reported their sex as “other” or preferred not to declare it were not included because these categories were too small to estimate; “participant sex” denotes the respondent’s sex, not the gender of the depicted physician. Relative to the casual-clothing image at position 1, the adjusted odds of acceptance were higher for each of the 4 later images (all p < 0.001), with the largest contrast observed for the scrubs-with-white-coat image at position 5 (OR, 39.10; 95% CI, 13.11-116.66; Table 5). Because acceptance approached 100% at the later positions, the probability scale is more interpretable: adjusted acceptance increased from 57.6% to 97.8% at position 5, a risk difference of +40.3 percentage points (cluster-bootstrap 95% CI, 32.6-48.0; full results in Supplementary Table 7). No participant characteristic reached significance in this model, although postgraduate education showed a borderline association (p = 0.051; Table 5). The direction and large magnitude of the position-based image contrasts were broadly robust across alternative estimators (independence working correlation, Firth penalized likelihood, and random-intercept model; Supplementary Table 7), although Firth does not account for participant clustering and serves only as a limited sensitivity check for sparse-data and near-separation behavior, rather than as a clustering-adjusted alternative to the GEE.

TABLE 5. Multivariable Model of Willingness to Proceed with Treatment Across the Five Images in the Fixed Attire Sequence, Male-Image Stream (915 Observations from 183 Participants).

Variable	Adjusted OR (95% CI)	p
Position 2: suit (vs. position 1: casual)	1.90 (1.44-2.50)	< 0.001
Position 3: suit + coat (vs. position 1: casual)	14.94 (7.56-29.52)	< 0.001
Position 4: scrubs (vs. position 1: casual)	21.84 (9.49-50.25)	< 0.001
Position 5: scrubs + coat (vs. position 1: casual)	39.10 (13.11-116.66)	< 0.001
Age (per 10 years)	0.92 (0.74-1.14)	0.451
Participant sex: female (vs. male)	2.08 (0.82-5.27)	0.121
Secondary education (vs. primary)	1.72 (0.70-4.24)	0.237
University (vs. primary)	2.75 (0.86-8.78)	0.087
Postgraduate (vs. primary)	4.07 (0.99-16.67)	0.051

Subgroup analysis by prior urological surgery

The association across the attire sequence was present regardless of previous surgical experience (exploratory Global Perception Score, male-image stream): the increase from casual clothing to scrubs with a white coat was similar among participants with (n = 85) and without (n = 104) prior urological surgery (Friedman p < 0.001 in both; Mann-Whitney p = 0.704 for the difference in magnitude).

Sensitivity analysis for presentation order

Because all images were shown in the same fixed order, with attire images in positions 1-5 and context images in positions 6-10, presentation position and attire category are completely confounded, and this cannot be resolved analytically. As a partial, imperfect check, we examined whether ratings continued to increase across positions 6-10, as would be expected if a general position effect (anchoring, learning, or acquiescence) drove the increase across positions 1-5. Trust showed no significant trend across these later positions (Page’s z = 0.75, one-sided p = 0.227, uncorrected; mean change ≤ 0.17 points), although the exploratory composite showed a small trend reaching nominal significance (z = 2.04, p = 0.021, uncorrected); ratings decreased slightly from positions 5 to 6 (4.26 to 4.03; p < 0.001) and were largely flat thereafter. However, because the context images differ from the attire images in content, equipment, and framing as well as in position, this check cannot isolate an order effect from a content-domain effect and does not establish that the attire-sequence association is free of order effects; it is reported as a partial, exploratory observation only.

DISCUSSION

In this image-based survey of 220 adult urological patients, ratings varied systematically across the fixed, ordered sequence of images shown, with the largest and most consistent differences appearing across the 5 images intended to vary in physician attire. Because the attire category was perfectly confounded with presentation position, these differences cannot be attributed to attire independently of order. Ratings of trust, choice likelihood, competence, professionalism, and related domains increased from casual clothing to surgical scrubs with a white coat, and willingness to proceed with treatment increased from slightly more than half of the participants to almost all. In the smaller female-image stream, which was analyzed separately, ratings varied across its fixed sequence. Because that stream was not randomized and differed systematically in size and composition from the male-image stream, no comparison between the 2 streams is made.

These findings are consistent with evidence that patients report greater trust in physicians who appear professionally dressed⁵ and with surveys showing clear expectations regarding physician dress.⁶ This extends to urology, a procedural specialty that has received comparatively little prior attention.^{10, 12} Within the fixed sequence used here, the images depicting surgical scrubs, alone or with a white coat, received ratings at least as high as those depicting business attire, consistent with reports that scrubs are increasingly accepted in procedural settings⁸ and with an international comparison in which scrubs were preferred specifically among Italian patients.⁹ Ratings were also higher for the white-coat images than for the corresponding earlier non-coat images, consistent with the symbolic role attributed to the white coat as a marker of medical identity⁴ and with evidence that impressions form rapidly from minimal visual input.² These descriptive patterns are consistent with the previous attire literature, but the present design cannot separate clothing from presentation position.

The near-absence of an association with the displayed surgical context warrants caution, given the order and stimulus-composition issues noted above and below. As detailed in the Methods section, the context images differed from one another not only in the operative setting but also in attire details, framing, and body orientation—most markedly in the robotic-console scene, which showed the physician seated and in profile rather than standing and facing the camera. Representing each scene recognizably arguably required some differences, but comparisons among them reflect these composite scenes rather than an isolated manipulation of surgical context; the near-null result should not be interpreted as evidence that the operative setting itself has no bearing on patient perceptions. If genuine, part of the explanation may be psychological: attire is an interpersonal cue processed rapidly and requires no specialized knowledge, whereas appraising a robotic console or laparoscopic tower requires prior understanding of what those platforms do,^{13, 14} and in the absence of such knowledge, participants may default to the person rather than the equipment.

The large adjusted OR for image 5 versus image 1 (Table 5) represents a position-specific stimulus contrast, not a generalizable attire effect. Within each image stream, each condition was represented by one rendering of a fixed physician identity; position, condition, and stimulus exemplar are therefore inseparable. The direction and overall large magnitude of the contrasts across the fixed attire-image sequence were broadly robust across alternative estimators, including an independence working correlation, Firth penalized likelihood, and a random-intercept model. The Firth analysis, however, does not account for participant clustering and serves only as a limited sensitivity check for sparse-data and near-separation behavior, rather than as a clustering-adjusted alternative to the GEE. Ratings did not continue to increase across the 5 subsequent context-image positions; however, this observation cannot be used to discount an order effect because those images differ from the attire images in content as well as position. An order effect within the attire sequence therefore remains unresolved.

Clinically, physician dress remains modifiable, and the pattern observed here—more favorable initial impressions toward the end of the attire sequence (scrubs, particularly with a white coat) than toward its beginning (casual clothing)—is consistent with that literature, although it cannot be isolated from image order. Because attire preferences vary by setting and country,⁷ because these data describe impressions of static images, and given the limitations noted above, the observations are best regarded as hypothesis-generating rather than as a basis for changing dress policy. Specialty-specific surveys in ophthalmology, emergency medicine, and obstetrics and gynecology have similarly linked attire to patient impressions.^{11, 16, 17}

Several limitations qualify these findings. First, and most importantly, all participants viewed the images in the same fixed order, from casual clothing to scrubs with a white coat; presentation position is therefore completely confounded with attire category, and results should be interpreted as an association across this ordered sequence, rather than as an attire-specific effect. The subsequent context-image positions do not adequately check this confound because those images also differ in content, equipment, and framing; only randomizing image order in a future study would resolve this issue. Second, attire and context were varied in separate, non-factorial image sets, so the 2 factors cannot be compared or tested for interaction; the compressed context-arm distribution further limits the interpretation of its small effect sizes. Third, within each image stream, 1 fixed AI-generated physician identity was used, with a single rendering per condition. Therefore, CIs are conditional on these particular images and may overstate the precision of categories such as “casual clothing” or “robotic context.” The context images also differed incidentally beyond the operative setting. A design sampling several exemplars per condition, with stimulus treated as a random effect, would be needed. Fourth, the perception items were highly redundant and should not be considered independent evidence. Fifth, image gender followed participant preference rather than randomization, so no inference about physician-gender is possible. Sixth, the instrument was purpose-built and unvalidated, health literacy was assessed using a single non-validated item, image review was internal, and no sample-size calculation was performed; non-significant findings (especially those from the small female-image stream) are not evidence of equivalence. Finally, participants were a convenience sample from urology clinics in 1 country, decliners were not recorded, and responses to images may not reflect behavior during real consultations.

In conclusion, ratings varied systematically across the fixed, ordered image sequence used in this survey, most consistently across the images intended to vary in physician attire. Because the attire category was completely confounded with presentation position, this association cannot be attributed to attire independently of order. Associations with the displayed surgical context were smaller and are further limited by unintended differences among those images. Prospective, order-randomized studies are needed to determine whether these associations reflect physician attire itself and whether they translate into routine clinical encounters before informing dress policies.

Supplementary Materials

Supplementary Figures

https://balkanmedicaljournal.org/img/files/7-15-supplement-figures.pdf

Supplementary Materials and Methods

https://balkanmedicaljournal.org/img/files/7-15-supplement-material.pdf

Supplementary Questionnaire

https://balkanmedicaljournal.org/img/files/7-15-supplement-question.pdf

Supplementary Tables

https://balkanmedicaljournal.org/img/files/7-15-supplement-table.pdf