A 37-question instrument was developed de novo combining multiple-choice questions and numeric free-text responses to capture the study objectives (see Multimedia Appendix 1 for the survey instrument). The instrument was revised and pilot-tested among the study team for functionality prior to distribution. Initial questions screened for respondents who had any active or prior health care–related work experience. Further questions collected baseline demographics. Demographic information included age, sex, race, insurance status, and educational level. Health literacy was measured by polling the frequency of needing assistance in understanding health material. Prior knee-related history (history of knee pain, knee MRI, knee PT, or knee surgery) and general risk tolerance were also collected.

Subsequent sections presented randomized scenarios of a fixed knee injury and brief communications with a physician and then collected respondent preferences, expectations, and satisfaction regarding the management of this knee injury. Education regarding the meniscus and MRI function was provided to participants alongside the standardized vignette before formal questioning began.

A nonclinical sample was selected to isolate how patient-facing portal wording shapes initial preferences before specialist counseling, reducing confounding from an orthopedic clinical sample that may already be exposed to specialist counseling and, thereby, may influence responses.

Participants were recruited through Amazon Mechanical Turk (MTurk), a clinically validated online crowdsourcing marketplace tool in which “requestors” financially compensate “workers” for completing human intelligence tasks (HITs), which commonly include health-related survey-based studies [13]. To enhance the quality of respondents, participation was restricted to workers with a prespecified HIT rating demonstrating a history of high-quality responses (≥95%). Inclusion criteria were age of 18 years or above, residence in the United States, and an MTurk HIT rating of 95% or higher. These criteria were enforced using MTurk worker qualifications. Exclusion criteria were respondents indicating any current or prior health care–related work experience and failure to pass embedded attention checks throughout the survey. Health care work experience was excluded to reduce bias from existing medical knowledge that could alter the reported outcomes. Respondents who failed any attention check were excluded to improve data quality. To avoid missing data, all survey questions required a response to advance and submit.

All survey questions were administered using an online survey platform (Qualtrics XM; Qualtrics International Inc) with embedded security checks enabled. The survey required approximately 3 to 10 minutes to complete. At the end of the Qualtrics survey, each participant was presented with a completion code, which they were required to enter into the MTurk interface to receive compensation. Completion codes were matched between Qualtrics and MTurk to ensure that each response corresponded to a unique, valid respondent. Incomplete survey responses or entries with invalid codes were rejected and did not receive compensation.

All participants were presented with the same standardized clinical vignette:

MRI impressions were identical for all participants to hold pathology constant and simulate real-world patient access to MRI results in patient portal systems. Following this, participants were randomized in a 1:1:1 ratio into 3 separate framing groups: neutral, degenerative, and damage. After group randomization, participants were presented with a brief simulated portal message from the joint specialist using language specific to the assigned framing groups (Figure 1). Following this, participants’ preferences, expectations, and satisfaction were assessed.

The primary outcomes were preferred next steps in management and perceived level of confidence in conservative therapy success. This was assessed by presenting a multiple-choice question of conservative (nonoperative or second opinion) and invasive (procedural) options: (1) start PT, (2) steroid injection, (3) rest and reassess, (4) surgery, and (5) seek a second opinion. Invasive options were surgery and steroid injection. Conservative options were PT, rest and reassess, and seeking a second opinion.

The secondary outcome was retained satisfaction in a follow-up scenario in which the joint specialist disagreed with the respondents’ preferred next step in management and whether satisfaction had any association with a respondent’s initial preferences. Satisfaction was measured on an ordinal response scale (“very dissatisfied,” “dissatisfied,” “neutral,” “satisfied,” and “very satisfied”).

Descriptive statistics were presented as means and SDs. Categorical variables, including Likert distributions, were compared using chi-square and Fisher exact tests. Pairwise comparisons were reported using odds ratios (ORs) with 95% CIs. Multivariable logistic regression analysis was performed to assess whether associations between framing groups persisted after adjusting for baseline characteristics. A P value of .05 or less was considered statistically significant. Given the exploratory nature of this study and the limited sample size, formal correction for multiple comparisons was not performed. Secondary pairwise comparisons and post hoc stratified analyses were considered hypothesis generating and should be interpreted cautiously.

Exploratory post hoc sensitivity analyses were performed stratifying by prior knee MRI status. In each stratum, multivariable logistic regression models were used for preference for invasive management and confidence in PT. Formal interaction testing was used to assess whether prior knee MRI status carried associations between framing groups and outcomes.

Per institutional policy, this study was determined not to constitute human subjects research and therefore did not require formal institutional review board review (Multimedia Appendix 2). All survey responses were anonymous and contained no identifying information. Participants received US $0.50 in compensation upon completion.

In total, 266 American adults completed the survey. Of these 266 participants, 67 (25.2%) were excluded due to indicating prior or current health care–related work experience, and 4 (1.5%) were excluded due to failing the embedded attention checks. No data were missing as all survey questions required a response. A total of 195 respondents were included for analysis: the neutral framing group included 67 (34.4%) respondents, the degenerative framing group included 63 (32.3%) respondents, and the damage framing group included 65 (33.3%) respondents.

Participant characteristics were highly similar across groups (Table 1). Mean respondent age was approximately the same across all 3 groups (neutral: 33.3, SD 4.4 years; degenerative: 35.0, SD 6.3 years; damage: 33.5, SD 3.8 years). All 3 groups were predominantly male (155/195, 79.5%) and White (187/195, 95.9%), with the vast majority holding a bachelor’s degree or higher (187/195, 95.9%). Prior knee-related experiences (prior knee pain of >4 weeks, prior knee MRI, prior knee PT, or prior knee surgery) were comparable across groups, and each group scored relatively similarly with regard to self-reported risk tolerance (P=.13; Table 1).

Following the presentation of the physician portal message, respondents were polled on their preferred next steps in management. The groups varied significantly in the overall distribution of their treatment preferences (P=.047). The damage group had the highest rate of preference for invasive management (ie, surgery or intra-articular injection; 48/65, 73.8%) compared to the neutral (36/67, 53.7%) and degenerative (37/63, 58.7%) groups and was significantly more likely to prefer invasive management compared to the neutral group (OR 2.43, 95% CI 1.17-5.06; P=.02). The neutral group was most likely to seek conservative management (ie, rest and reassess, PT, or seeking a second opinion; 31/67, 46.3%), followed by the degenerative group (26/63, 41.3%) and damage group (17/65, 26.2%; Table 2).

In multivariable logistic regression analysis, the damage group remained independently associated with greater odds of preferring invasive management compared with the neutral group (OR 2.34, 95% CI 1.10-4.96; P=.03), whereas the degenerative group was not significantly associated with invasive preference (OR 1.24, 95% CI 0.60-2.55; P=.56; Figure 2).

Between-group expectations regarding confidence in conservative therapy success varied significantly (χ²₄=12.3; P=.02). In the damage group, respondents were most pessimistic about conservative therapy and more frequently indicated “little improvement” expected from PT (16/65, 24.6%) compared to the neutral (7/67, 10.4%) and degenerative (9/63, 14.3%) groups, whereas the neutral and degenerative groups more frequently indicated “moderate” and “great” improvement with PT over the damage group (neutral: 60/67, 89.6%; degenerative: 54/63, 85.7%; damage: 49/65, 75.4%; Table 3).

In multivariable logistic regression analysis for expectations regarding improvement with PT, respondents in the damage group had greater odds of expecting little improvement with PT compared with the neutral group (OR 3.21, 95% CI 1.19-8.70; P=.02), whereas the degenerative group did not differ significantly from the neutral group (OR 1.71, 95% CI 0.58-5.02; P=.33; Figure 3).

In exploratory post hoc analyses stratifying by prior knee MRI status, the observed direction of the framing effect on the primary outcomes remained similar across strata. Among respondents with prior knee MRI (92/195, 47.2%), damage framing was significantly associated with greater odds of preferring invasive management (OR 3.17, 95% CI 1.04-9.67; P=.04), whereas degenerative framing was not (OR 1.41, 95% CI 0.49-4.09; P=.52). Damage framing was also significantly associated with greater odds of expecting little improvement with PT (OR 5.14, 95% CI 1.16-22.80; P=.03), whereas degenerative framing was not (OR 2.30, 95% CI 0.46-11.55; P=.31).

Among respondents without prior knee MRI (103/195, 52.8%), primary outcome associations by framing group were directionally similar, but damage framing did not reach statistical significance for invasive management preference (OR 1.84, 95% CI 0.61-5.56; P=.28) or expectations regarding PT (OR 1.91, 95% CI 0.46-7.95; P=.37). On formal interaction testing, no significant effect modification by prior knee MRI status was present for invasive management preference (P=.76) or expectations regarding improvement with PT (P=.72).

When presented with the scenario in which the treating physician disagreed with the respondent’s preferred management, overall physician satisfaction ratings did not differ significantly between groups (χ²₆=6.7; P=.35).

While satisfaction under treatment discordance did not differ across groups, it did vary significantly by respondents’ preferred next step (χ²₁₂=23.0; P=.03). This finding was also consistent when satisfaction was dichotomized as “satisfied” or “very satisfied” vs not satisfied (χ²₄=11.7; P=.02). Of all respondents, only 40% (26/65) of those who preferred steroid injection indicated that they would retain any form of satisfaction with their joint specialist under disagreement. This was contrary to the retained satisfaction rates of those who preferred surgery (38/56, 67.9%), PT (26/40, 65%), second opinion (10/16, 62.5%), and rest and reassess (11/18, 61.1%; Table 4).

In understanding the discrepancy in retained satisfaction rates between those who preferred steroid injection and those who preferred surgery, 2 invasive options, between-group analysis of primary drivers of treatment selection showed that those who indicated a preference for steroid injection significantly weighed the speed of pain relief and return to normalcy as the most important drivers of their treatment selection compared to those who preferred surgery (steroid injection: 45/65, 69.2%; surgery: 28/56, 50%; P=.04; OR 2.25, 95% CI 1.07-4.73; Table 5). No other significant associations between drivers and treatment preferences were observed.

Our findings also complement broader research emphasizing the role of communication in shared decision-making. Decision aids and structured communication strategies have been shown to support patient engagement by presenting options in a balanced way and facilitating deliberation. Such tools aim to reduce unintended framing bias and help patients understand the implications of potential outcomes [20]. This aligns with the principle that framing should be used carefully to support, not skew, patient decisions.

These implications are particularly relevant in modern orthopedic practice, where patients increasingly access imaging reports and brief physician messages through electronic health portals before in-person discussion. Early exposure to framed language may shape expectations before shared decision-making formally begins, potentially influencing how patients interpret subsequent recommendations. Our findings suggest that early exposure to subtly pathologizing language may prime patients to view their condition as more severe, develop lower confidence in conservative care, and form preferences for invasive intervention before shared decision-making formally begins. Additionally, given ongoing concerns regarding overuse of invasive procedures for degenerative meniscal pathology, communication framing may represent an underrecognized contributor to patient demand for low-value interventions. Even modest shifts in patient expectations driven by language could have meaningful downstream effects on use at scale.

This study has several limitations that should be considered. First, the sample was limited to lay adults in the United States and may not represent the behaviors of orthopedic clinic populations in other settings; respondents were located in the United States and predominantly male, White individuals, and highly educated. As a result, the observed findings may not reflect real-world clinical populations both in the United States and elsewhere, limiting the generalizability of these findings. Second, management preferences were assessed using a hypothetical scenario rather than lived clinical experiences, and decision-making may differ when individuals are personally affected by a condition. Third, the magnitude of influence on patient behaviors that portal-style messaging possesses may vary depending on how a specific message is framed and could have many different impacts on attitudes and behaviors that fall outside the scope of this study. Fourth, while the exclusion of respondents who indicated any form of health care experience was done to increase internal validity, it may have also reduced generalizability by producing a more curated lay sample than would be expected in real-world clinical practice. Fifth, nearly half (92/195, 47.2%) of the respondents reported a prior knee MRI despite the relatively younger age of the study population. The plausible explanations for this are hypothesis generating in nature and could reflect selection bias in which participants with prior musculoskeletal history were drawn to the title and description of the survey on the MTurk web interface, limiting generalizability and necessitating a post hoc sensitivity analysis stratified by prior knee MRI status. Sixth, this study was designed around a hypothetical and simulated clinical vignette of a patient aged 60 years through which the study intervention was presented to participants who were relatively young. Therefore, the observed findings may not be reflective of the real-world decision-making of actual middle-aged and older patients, who would be most likely to experience this pathology, and may not capture the views of older orthopedic patients, who may possibly respond differently to the framing effects and treatment preferences. Seventh, while the primary aim of this study was to assess framing effect on invasive vs conservative treatment preferences, collapsing steroid injection and surgery into 1 category could have obscured additional behavioral and decision-making distinctions. Eighth, as exclusions were made after recruitment through health care experience screening and failed attention checks, subgroup sizes for each framing type, while similar in size, were not exactly even, and this could have influenced overall balance. Ninth, the language used in the damage framing group could reasonably have influenced the observed findings as describing “damage to the knee’s cartilage cushion” may have introduced both a pathologizing word (“damage”) and a structural association, possibly influencing perceived severity. Finally, although the study was randomized, the individual group sizes could have limited statistical power to detect additional smaller effects.

Accordingly, future research is needed to validate these findings, which are exploratory and hypothetical in nature, ideally replicating this study using a real-world clinical patient population. Further work should also seek to qualitatively characterize how patients interpret commonly used orthopedic terms such as “tear,” “degeneration,” and “damage,” which may inform the development of clearer, more patient-centered communication strategies.

In this exploratory cross-sectional study, subtle differences in physician portal message framing regarding an MRI impression of a degenerative meniscus tear were associated with shifts in treatment expectations and, in some cases, increased preferences for more aggressive management and decreased confidence in guideline-concordant conservative therapy. These findings suggest that physician portal language may influence hypothetical patient expectations and treatment preferences prior to clinical counseling. Future studies in real-world orthopedic populations are needed to determine whether these observed framing effects translate to actual treatment decisions.