Understanding Discordant Perceptions of Disease Severity Between Physicians and Patients With Eczema and Psoriasis Using Structural Equation Modeling

Key Points

Question

Why do patients and physicians assess severity differently, and why are some patients more symptomatic despite the same objective disease?

Findings

In this cross-sectional study of 1053 pairs of patients with psoriasis or eczema and physicians, 487 pairs (46.3%) demonstrated discordance; physicians focused on visually objective measures, while patients heavily considered the physical, functional, and emotional consequences of disease. These disease consequences in turn differed based on the individual’s resilience, self-efficacy, and tendency for negative social comparisons; physicians may also trivialize milder cases because of frequent exposure to severe diseases.

Meaning

These identified factors are potential targets for cognitive-behavioral interventions to bridge differing perceptions between patients and physicians.

Abstract

Importance

Patients and physicians often have differing opinions on the patient’s disease severity. This phenomenon, termed discordant severity grading (DSG), hinders the patient-physician relationship and is a source of frustration.

Objective

To test and validate a model explaining the cognitive, behavioral, and disease factors associated with DSG.

Design, Setting, and Participants

A qualitative study was first performed to derive a theoretical model. In this subsequent prospective cross-sectional quantitative study, the qualitatively derived theoretical model was validated using structural equation modeling (SEM). Recruitment was conducted between October 2021 and September 2022. This was a multicenter study in 3 Singapore outpatient tertiary dermatological centers. Dermatology patients and their attending physicians were recruited by convenience sampling. Patients were aged 18 to 99 years with psoriasis or eczema of at least 3 months’ duration and recruited only once. The data were analyzed between October 2022 to May 2023.

Main Outcomes and Measures

The outcome was the difference between global disease severity (0-10 numerical rating scale with a higher score indicating greater severity) as independently scored by the patient and the dermatologist. Positive discordance was defined as patient-graded severity more than 2 points higher (graded more severely) than physicians, and negative discordance if more than 2 points lower than physicians. Confirmatory factor analysis followed by SEM was used to assess the associations between preidentified patient, physician, and disease factors with the difference in severity grading.

Results

Of the 1053 patients (mean [SD] age, 43.5 [17.5] years), a total of 579 (55.0%) patients were male, 802 (76.2%) had eczema, and 251 (23.8%) had psoriasis. Of 44 physicians recruited, 20 (45.5%) were male, 24 (54.5%) were aged between 31 and 40 years, 20 were senior residents or fellows, and 14 were consultants or attending physicians. The median (IQR) number of patients recruited per physician was 5 (2-18) patients. Of 1053 patient-physician pairs, 487 pairs (46.3%) demonstrated discordance (positive, 447 [42.4%]; negative, 40 [3.8%]). Agreement between patient and physician rating was poor (intraclass correlation, 0.27). The SEM analyses showed that positive discordance was associated with higher symptom expression (standardized coefficient B = 0.12; P = .02) and greater quality-of-life impairment (B = 0.31; P < .001), but not patient or physician demographics. A higher quality-of-life impairment was in turn associated with lower resilience and stability (B = −0.23; P < .001), increased negative social comparisons (B = 0.45; P < .001), lower self-efficacy (B = −0.11; P = .02), increased disease cyclicity (B = 0.47; P < .001), and greater expectation of chronicity (B = 0.18; P < .001). The model was well-fitted (Tucker-Lewis: 0.94; Root Mean Square Error of Approximation: 0.034).

Conclusions and Relevance

This cross-sectional study identified various modifiable contributory factors to DSG, increased understanding of the phenomenon, and set a framework for targeted interventions to bridge this discordance.

This cross-sectional study tests and validates a model explaining the cognitive, behavioral, and disease factors associated with discordant severity grading between physicians and patients with eczema and psoriasis.

Introduction

Symptom burden from a disease varies widely between patients, and the evaluation of severity also differs between patients and their physicians.¹ We use the term discordant severity grading (DSG) to represent this concept where the patient and physician each arrive at a different assessment of disease severity.

Despite its frequent occurrence, this phenomenon is poorly understood. Studies on DSG have found a prevalence ranging from 22% to 43% with low patient-physician concordance.^{2,3,4,5,6,7,8} Most report a tendency for patients to report higher disease severity relative to the physician.^2,3,4,5,6 Previous studies focus heavily on demographic and disease factors, reporting a tendency for female patients,⁵ older patients,^9,10,11,12 and those with concurrent psychiatric disorders to report higher severity.^2,4,6,8 Inconsistent associations were seen with education level,^2,5 initial quality of life (QOL),^3,5 and disease severity.^2,3

The reasons behind this phenomenon are likely complex, and past studies using only quantitative approaches may not adequately explore the interrelated psychological and behavioral constructs.¹ These studies are also bound by predetermined theories of discordance and preselected variables.^3,12 For example, there has been minimal investigation into the physician factors that contribute to this phenomenon.

To address these gaps, we used a sequential exploratory mixed-methods study beginning with a qualitative study in 2020 that involved in-depth 1-on-1 interviews with 18 patients and 12 dermatology health care physicians.¹³ Using grounded theory as a qualitative approach, we identified new factors and associations, which were used to derive a comprehensive model explaining DSG. In this model, differences in scoring stem from differing perspectives. While physicians heavily emphasize the visible physical signs and biomedical aspects of disease, patients focus on the functional and emotional impact.^1,13 The degree of QOL impairment is moderated by the patient’s resilience and coping abilities. Factors such as self-consciousness, need to emphasize symptoms, underlying motives, and the context of the consultation (eg, physician trust, empathy, burnout, and strategies to reduce discordance) are associated with the reporting of symptoms to physicians. Concurrently, patients and physicians have different experiential perspectives of disease, with physicians comparing the patient’s severity relative to other patients, while patients may compare themselves with their peers and social circle.

In the current study, we first assess the prevalence and direction of DSG (whether patients or physicians tend to assign a higher severity). Next, using structural equation modeling (SEM), we validate the qualitatively derived model with quantitative data. The SEM method allows us to test the multidimensional constructs and complex associations that typify this phenomenon. We hypothesize that this derived model will demonstrate a good fit with clinical data, and the psychobehavioral and cognitive constructs previously identified will be significantly associated with DSG. Understanding the drivers of discordance is a necessary first step toward building a patient-centered care model.

Methods

Study Participants

In this cross-sectional study, patients and physicians were recruited from 3 outpatient tertiary academic dermatological centers in Singapore (National Skin Centre, National University Hospital, and Alexandra Hospital) using convenience sampling. Patients were included if they were aged 18 to 99 years, diagnosed with psoriasis or eczema with a disease duration of at least 3 months, and able to complete a self-administered survey in English. Inclusion criteria for physicians included at least 2 months of dermatological experience.

Patients and physicians were asked to complete the questionnaire independently after a consultation. Each patient contributed once to the data, but physicians could contribute more than once (ie, may complete surveys for multiple different patient encounters). Participants were reimbursed for their time. All participants provided written informed consent, and the study was approved by the National Healthcare Group institutional review board (reference number 2021/00110). The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline was followed.

Model and Survey Development

Because of limited data and the lack of an existing framework, we first developed a qualitative model¹³ as described previously and used that as the basis for this study. To operationalize and measure the earlier identified themes, a scoping review of the literature was done to identify validated scales that best represented the constructs of interest. When validated tools were not available, eg, degree of discordance and the tendency to compare, ad hoc items and response scales were developed by the study team. The development and selection of the final questions (eFile 1 in Supplement 1) were refined through discussion among team members.

Instruments used in the patient’s questionnaire included: QOL (Skindex-16),¹⁴ anxiety (Generalized Anxiety Disorder-7),¹⁵ various domains of illness perception (Revised Illness Perception Questionnaire [IPQ-R]),¹⁶ resilience (Brief Resilience Scale),¹⁷ personality (Ten Item Personality Measure [TIPI]),¹⁸ physician empathy (Jefferson Scale of Patient Perceptions of Physician Empathy [JSPPPE]),¹⁹ and trust in the physician (Wake Forest Trust Scale).²⁰ Measures used in the physician’s questionnaire included the Difficult Doctor Patient Relationship Questionnaire (DDPRQ-10).^21,22 Objective severity was assessed using body surface area (BSA) and investigator global assessment (IGA) or physician global assessment (PGA) for eczema and psoriasis, respectively.^23,24,25 The IGA for eczema and PGA for psoriasis assessed severity based on observable characteristics such as the degree of erythema, papulation, thickness, and scaling of the rash. Patient demographics (age, sex, race and ethnicity, education level, paying class), visit type (first visit or follow-up), and physician demographics (age, sex, race and ethnicity, rank) were collected and adjusted for as control variables.

The primary outcome was the difference in disease severity grading between patients and physicians. This was based on the question, “How severely would you grade your/the patient’s skin condition?” as an overall global assessment, scored on a 0 to 10 numerical rating scale with a higher score indicating greater severity. This outcome was subsequently categorized according to whether it was positive (patient graded 2 or more points higher than the physician) or negative (patient graded 2 or more points lower than the physician). The decision to categorize the outcome was based on earlier findings showing that positive and negative discordance are likely distinct constructs as opposed to extreme anchors on a linear scale.¹ Intraclass correlation (ICC) was also used to assess the level of agreement between patient and physician reported severity.

Because of the lack of an existing reference standard, we established the cutoff of 2 points through 2 main methods. First, we used a statistical distributional method of half an SD as a minimally important difference,^26,27 translating to a cutoff of more than 2, based on the SD of 2.6. Second, we used an opinion (expert)–based approach by considering cutoffs used in previous studies of discordance. In a previous systematic review, a cutoff of 2 was the most frequently used.¹

The questionnaire was piloted in 15 patient-physician pairs in the National Skin Centre and National University Hospital between July 2021 and August 2021 for readability, ease of completion, and feasibility. Refinements were made before the recruitment of the main cohort in October 2021.

Statistical Analysis

We used 2 estimation methods for sample size calculation. The first method was a rule-of-thumb calculation using a 10 to 20 case per parameter ratio for SEM,²⁸ giving a sample of 500 to 1000 for a proposed model with 50 parameters. The second was an estimation based on 6 latent factors, 3 variables per factor, and a communality between 0.2 and 0.8, which yielded a sample size estimation of 700 for good and 1600 for excellent agreement.²⁹ From this, we targeted the recruitment of 1000 patient-physician pairs.

Instrument reliability was assessed using Cronbach α. Confirmative factor analysis was used to assess the validity of the proposed latent variables and to ensure the adopted questionnaires apply to the present study’s sample. Subsequently, a 1-step SEM (R package Lavaan³⁰) was used to assess the associations between latent variables and the outcomes of interest (positive discordance vs concordance; negative discordance vs concordance). Standardized estimates were calculated using the diagonally weighted least squares estimator. The models were adjusted for disease factors and patient and demographic characteristics.

Competing variations of the model were developed based on alternate theoretical justifications, including assessment of discordance as a continuous instead of categorical variable, using the ratio of patient to physician grading as an outcome, inclusion or exclusion of disease duration, and inclusion or exclusion of the context of the consultation. These models were compared using model-fitting indices, and the best-fitting and most parsimonious model was chosen.

Statistical analysis was carried out in Stata/SE 17.0 (StataCorp LLC), and the SEM modeling was implemented using the Lavaan package of R, version 4.1.2 (R Project for Statistical Computing). Statistical tests were all 2-sided with a significance level of P < .05.

Results

A total of 44 physicians and 1057 of their patients were recruited between October 2021 and September 2022 (Table 1). Four patients withdrew from the study (reasons included insufficient time to complete, personal preference, and 2 who were younger than 18 years old), leaving 1053 pairs available for analysis. Of the 1053 patients (mean [SD] age, 43.5 [17.5] years), a total of 579 (55.0%) patients were male, 802 (76.2%) patients had eczema, and 251 (23.8%) had psoriasis. Of 44 physicians recruited, 20 (45.5%) were male, 24 (54.5%) were aged between 31 and 40 years, 20 were senior residents or fellows, and 14 were consultants or attending physicians. The median (IQR) number of patients recruited per physician was 5 (2-18) patients. The data were near complete with less than 0.01% missing data points. Descriptive results of the questionnaire responses, stratified by disease condition, are summarized in eFile 2 in Supplement 1. Patients with missing points were automatically excluded from the SEM modeling.

Table 1. Characteristics of Recruited Patients and Physicians.

Characteristic	No. (%)
Patient demographics (n = 1053)
Age, mean (SD), y	43.5 (17.5)
Sex
Male	579 (55.0)
Female	474 (45.0)
Race and ethnicity
Chinese	780 (74.1)
Malay	116 (11.0)
Indian	94 (8.9)
Othera	63 (6.0)
Education level
Primary	33 (3.1)
Secondary	207 (19.7)
High school/junior college	391 (37.1)
Bachelor	340 (32.3)
Postgraduate	82 (7.8)
Marital status
Single	393 (37.3)
Married	509 (48.3)
Dating	87 (8.3)
Divorced/separated	49 (4.7)
Widowed	15 (1.4)
Paying status
Government subsidized	939 (89.2)
Self-paying	114 (10.8)
Recruitment No. per center
National Skin Centre	536 (50.9)
National University Hospital	480 (45.6)
Alexandra Hospital	37 (3.5)
Health care professional demographics (n = 44)
Age, y
20-30	8 (18.2)
31-40	24 (54.5)
41-50	8 (18.2)
>50	4 (9.1)
Sex
Male	20 (45.5)
Female	24 (54.5)
Rank
Medical office/junior resident	3 (6.8)
Senior resident/fellow	20 (45.5)
Associate consultant/consultantb	14 (31.8)
Senior consultantb	7 (15.9)
Disease characteristics and ratings
Disease condition
Eczema	802 (76.2)
Psoriasis	251 (23.8)
Duration of disease, mean (SD), y	32.2 (20.4)
Objective severity
Body surface area (physician-rated)
Median (IQR), %	2 (1-8)
Mean (SD), %	6.8 (12.1)
Global assessment (physician-rated)
IGA for eczema
0	34 (4.2)
1	211 (26.3)
2	314 (39.2)
3	223 (27.8)
4	20 (2.5)
PGA for psoriasis
0	9 (3.6)
1	65 (25.9)
2	109 (43.4)
3	60 (23.9)
4	8 (3.2)
Patient-graded severity (from 0-10, 10 being most severe), mean (SD)
Overall	5.6 (2.3)
Emotional	5.2 (2.9)
Functional	4.5 (3.0)
Physical	6.0 (2.7)
Physician-graded severity (from 0-10, 10 being most severe), mean (SD)
Overall	3.6 (2.2)
Emotional	4.1 (2.4)
Functional	3.7 (2.4)
Physical	4.0 (2.4)
Discordance in severity grading
Mean difference, mean patient rating − mean physician rating (SD)	2.0 (2.6)
Proportion discordant (>2-point difference between patient and physician)
Positive discordant	447 (42.4)
Negative discordant	40 (3.8)
Concordant	566 (53.8)

Abbreviations: IGA, investigator global assessment; PGA, physician global assessment.

^a Other refers to any race or ethnicity not in the main 3 racial categories in Singapore, including Eurasian and Arab populations.

^b Associate consultants, consultants, and senior consultants are the equivalent of attending physicians in the US.

Patients tended to grade severity higher than physicians, with a mean (SD) difference of 2.0 (2.6) points. A total of 447 (42.4%) were positively discordant (patient graded worse severity than physician), 40 (3.8%) were negatively discordant, and 566 (53.8%) were concordant. Agreement between patient-graded and physician-graded severity was poor, with an ICC of 0.27.

Cronbach α showed generally good reliability of included measures (α, 0.70-0.95) except for stability from the personality measure TIPI (α, 0.39). The confirmative factor analysis showed good structural validity of the proposed latent constructs, with a comparative fit index of between 0.91 and 1.00 (eFile 3 in Supplement 1).

As most (447 of 487) discordant pairs were positively discordant (patients graded disease more severely than physicians), this study focused primarily on the model for positive discordance (Figure 1).

Figure 1. Structural Equation Modeling Explaining Positive Discordant Severity Grading (DSG).

Latent constructs are represented by blue-gray ovals and observed variables by light gray rectangles. Arrows and values pointing from latent variables into observed variables reflect factor loadings (confirmatory factor analysis). Arrows and values pointing to quality-of-life (QOL) impairment or DSG represent path coefficients. Bidirectional arrows represent correlations between latent constructs. Nonsignificant variables are excluded from the figure for ease of presentation (full model in eFile 4 in Supplement 1). The model was adjusted for patient age, sex, race and ethnicity, education, paying class, disease condition; physician age, sex, race and ethnicity, rank; and physician perception of the patient’s symptom reporting.

^aP < .05.

Positively discordant grading was more likely if patients had greater QOL impairment (B = 0.31; P < .001), higher symptom expression (B = 0.12; P = .02), and lower objective disease severity (B = −1.21; P < .001). A higher QOL impairment was in turn associated with the latent constructs of resilience and stability (less stable personality, poorer resilience, higher anxiety; B = −0.23; P < .001), increased negative social comparisons (comparing with others, and feeling unhappy about one’s own situation, B = 0.45; P < .001) and lower self-efficacy (poorer control and understanding of the disease, B = −0.11; P = .02). A higher QOL impairment was also associated with increased disease cyclicity (B = 0.47; P < .001) and greater expectation of chronicity (B = 0.18; P < .001).

The perceived normality of having rashes and the frequency of encountering others with worse skin were not statistically significant. Patient and physician demographics were included in the model as control variables and were similarly nonsignificant. Of disease factors, patients with eczema or a lower objective severity were more likely to have positive discordance (eFile 4 in Supplement 1).

Model fitting indices showed good fit with a comparative fit index (CFI) of 0.75, Tucker-Lewis index of 0.94, root mean square error of approximation of 0.03, and standardized root mean square residue of 0.04.

The model was rerun with negative discordance as the primary outcome (negative discordant, 40; concordant, 566). The number of control variables was reduced to allow for model convergence. Self-efficacy and symptom expression were no longer statistically significant, although an increased frequency of encountering someone with worse rashes was negatively associated with QOL impairment (Figure 2). The significance of the remaining variables was unchanged.

Figure 2. Structural Equation Modeling Explaining Negative Discordant Severity Grading (DSG).

Latent constructs are represented by blue-gray ovals and observed variables by light gray rectangles. Arrows and values pointing from latent variables into observed variables reflect factor loadings (confirmatory factor analysis). Arrows and values pointing to quality-of-life (QOL) impairment or DSG represent path coefficients. Bidirectional arrows represent correlations between latent constructs. Nonsignificant variables are excluded from the figure for ease of presentation (full model in eFile 4 in Supplement 1). The model was adjusted for patient age, sex, education; physician sex; and physician perception of the patient’s symptom reporting.

^aP < .05.

Four other models were tested to examine the consistency of these findings and the plausibility of competing theories. These showed a poorer fit compared with the original (eFile 5 in Supplement 1). The alternate models included the inclusion of disease duration as a control variable (model 2), analysis of DSG as a continuous rather than categorical outcome (model 3), use of the patient-to-physician severity ratio as a measure of discordance (model 4), and inclusion of context of the consultation in the model (model 5). A detailed rationale for the primary and alternate models and statistical output are presented in eFile 4 in Supplement 1. Of note, the full model including the context of consultation and patient experience (model 5) could not be run due to insufficient cases for the number of parameters. The model could be fitted with a reduction of parameters, but as the model fit was poorer, this was excluded for a more parsimonious model. Nevertheless, results from these alternate models were largely consistent with the final selected model, with the same variables remaining statistically significant, adding credibility to the main findings.

Multilevel modeling accounting for clustering at the physician level found an association between positive discordance and lower rating of physician empathy (B = −1.51; P = .002; ie, patients graded their physicians less empathetic if they graded severity higher than the physician), and between negative discordance and lower trust in the physician (B = −0.67; P = .049; ie, patients had lower trust in their physicians if they graded severity lower than the physician). The remaining coefficients were nonsignificant (Table 2). As the outcome of discordance was dichotomous, we could not account for the clustering effect of physicians in the SEM model. However, the ICC quantifying the proportion of variance in outcome measure associated with the physician was low (0.11; 95% CI, 0.04-0.23), suggesting minimal clustering effect.

Table 2. Association of Discordant Severity Grading With the Patient-Physician Relationship^a.

Variables	Unstandardized coefficient (SE)	P value
Empathy
Concordance	0 [Reference]	NA
Positive discordance	−1.51 (0.49)	.002b
Negative discordance	−0.17 (0.19)	.38
Trust
Concordance	0 [Reference]	NA
Positive discordance	−0.60 (0.89)	.50
Negative discordance	−0.67 (0.34)	.049b
Satisfaction with explanation
Concordance	0 [Reference]	NA
Positive discordance	−0.12 (0.16)	.44
Negative discordance	−0.06 (0.06)	.32
Satisfaction with treatment
Concordance	0 [Reference]	NA
Positive discordance	−0.15 (0.18)	.42
Negative discordance	−0.13 (0.07)	.07

Abbreviation: NA, not applicable.

^a Multilevel analyses were adjusted for patient age, sex, race and ethnicity, education, paying class, and disease type, accounting for clustering at the physician level.

^b Indicates statistical significance at the P < .05 level.

Discussion

In this cross-sectional study, we were able to confirm the clinical validity and applicability of the present model explaining DSG. Supporting our prior understanding, higher patient-graded severity arose from greater QOL impairment, while higher physician-graded severity related to greater objective severity.^1,13 Most of the previously hypothesized factors for explaining discordance (resilience, negative social comparisons, self-efficacy, expectation of chronicity, QOL impairment, disease cyclicity, and symptom expression) were found to be statistically significant.

Critically, we found that patient and physician demographics were not significantly associated with discordance, supporting our belief that these may have been surrogates for untested cognitive-behavioral factors when found to be statistically significant in previous studies. The temptation to attribute outcomes to easily available demographic variables may lead to a misinterpretation of the phenomenon and impedes attempts to change the outcome because these demographics are typically nonmodifiable.

Measures of behavioral, cognitive, and psychological factors are not routinely collected and are thus less readily available. However, many of these factors are modifiable and provide a framework for subsequent intervention. For example, resilience and self-efficacy may both be strengthened through cognitive behavioral therapy, education, and counseling for stress management.^31,32,33

The present study highlights the association of negative social comparisons with QOL impairment and DSG. The way one evaluates his or her own state relative to others has the power to positively or negatively shape one’s mental health and health-seeking behavior.^34,35,36 By leveraging the human tendency to compare, carefully designed interventions to shape one’s evaluation of disease relative to others may consequently improve one’s perception and acceptance of the own skin condition.³⁷ Social media can be harnessed as a powerful tool to educate, engage, motivate, empower, and inspire patients to live their best lives.^38,39

Disease cyclicity is another important factor. Equipping patients with contingency and action plans, healthy coping mechanisms, and a realistic outlook can facilitate self-efficacy^40,41,42 and reduce the distress, helplessness, and intrusiveness^43,44 associated with diseases that run an unpredictable and cyclical nature.^45,46

It is equally important to acknowledge the crucial contribution of physicians to the phenomenon of discordance, rather than viewing it as a patient problem. The tendency for physicians to undergrade severity may arise from various factors, including daily exposure to severe diseases that leads one to perceive milder cases as trivial. Furthermore, medical training focuses on objective and physiological aspects of disease and often overlooks the patient’s subjective experience, such as the itch and functional impairment that even a small patch of rash can cause. This can result in a physician-focused or disease-focused approach to care that fails to meet the patient’s needs. It is important that we prioritize patient-focused outcomes and cultivate effective communication habits that allow us to elucidate values that are important to the patient.

Although the present study was limited to dermatology patients with eczema and psoriasis and may not generalize to other patients, we believe that these results may be relevant in many diseases, especially when the functional and emotional burden of disease outweighs the clinical signs. A high symptom burden has multifactorial origins, with objective severity being only one.^13,47,48 While physicians may view such patients as being not more ill, patients may view themselves as having worse health and being more disabled.⁴⁹ Focusing only on the biomedical aspects (eg, further diagnostic investigation or escalation of treatment) to resolve this discordance could lead to unnecessary costs and adverse effects.^50,51,52

Patient and Public Involvement

This study’s patient and public involvement (PPI) representatives included a patient with eczema and another with psoriasis. Representatives were first asked to provide insights on what they believed influenced the perception of symptom burden among patients and physicians (before reviewing preliminary models). As factors identified by PPI representatives were already in the preliminary model, and the proposed association between factors was consistent, no changes to the model were made. Subsequently, the SEM model with its coefficients was presented to participants to provide them an opportunity to provide feedback and propose changes. They agreed with the contents and provided qualitative comments and directions for future work (for instance, a representative had shared about the importance of social media as an avenue to explore in future studies as social media can contribute toward negative social comparisons, but can also be harnessed to improve illness coherence—much like a double-edged sword).

Limitations

Limitations of the study include sampling bias, as patients who were less cooperative or amicable may not have been asked or declined participation. There is also no existing validated scale for measuring or defining DSG, which led us to adopt the most used scale and cutoff from existing literature. Nevertheless, models using variations of the outcome showed similar results and support the analysis. Using the difference between patient and physician score and dichotomizing it has easier interpretability and was thus preferred as opposed to using a ratio, which has its metric advantages. We were unable to account for the clustering of observations at the physician level for the SEM analysis (the statistical package could not handle clustering for dichotomous outcomes). However, the ICC assessing the class effect of the physician was low, suggesting little clustering effect. While we sought to explore physician-related factors contributing to DSG, we did not identify any statistically significant factors, which may reflect an insufficient probing of physician beliefs and biases.

Conclusions

In this cross-sectional study, we found a high prevalence of discordance in patient-physician graded severity, with almost half of all encounters being discordant. As discordance is associated with poorer patient-physician relationships, patient dissatisfaction,⁵³ and poorer subsequent QOL,^4,7 it is important to recognize and understand the factors associated with this phenomenon. This study brings to the surface key lessons for dermatologists—that many reasons for discordance are, in fact, modifiable. This paves the way for further interventional work in this area and should be points of focus for the practicing physician. Further studies can assess longitudinal changes in disease grading over time, the effect of DSG on future health, and treatment response; explore deeper into physician factors; and evaluate different measures and metrics for assessing DSG.

Supplement 1.

eFile 1. Constructs assessed and their respective questions in the patient and physician survey, along with changes made during the pilot phase.

eFile 2. Descriptive statistics of patient related responses, stratified by disease type

eFile 3. Cronbach’s a and confirmatory factor analysis (CFA) loadings of the latent and observed constructs with their respective statistical output.

eFile 4. Statistical output of the 5 SEM models with the rationale for the design and selection of the primary and alternate models

eFile 5. Model fitting indices of selected and alternative models

Supplement 2.

Data Sharing Statement