An important dimension of medical education is the students' attitudes toward patients and how those attitudes may change over the course of the students' education. This is of particular importance regarding diagnoses that have been associated with negative attitudes, such as AIDS,1,2,3,4 psychiatric diagnoses,5,6,7,8 insulin-dependent diabetes in children,9 and chronic pelvic pain.10 Diagnoses such as these may become stigmatizing labels. Seeing certain medical conditions on a patient's chart can modify the mindset a clinician brings to the clinical encounter. For example, knowledge that a patient has a problem with alcohol has been shown to lower ratings of a patient's likability and attractiveness prior to the initial interview.8 Moreover, attitudes toward a diagnostic group influence clinical behavior, as demonstrated by Yedidia and co-workers, who found that physicians' negative attitudes toward patients with AIDS strongly predicted the physicians' pursuing less aggressive treatment and demonstrating more avoidant clinical behaviors with HIV-infected patients.3
It is important to study further the potency with which diagnostic descriptors may shape attitudes and influence clinical behaviors, and the effects curricular experiences may have. Optimal evaluation of a medical condition's effect on physicians' attitudes, however, requires an instrument that can assess responses to the diagnostic condition of interest and to control conditions. To the best of our knowledge, no such attitude scale exists. Examples of uses for such a scale include: (1) determining whether attitude shifts over time are specific to the target condition or due to broader factors affecting attitudes more generally, (2) investigating stigma by comparing a diagnosis hypothesized to be stigmatized with other diagnoses of equivalent medical complexity, and (3) rating the same condition with different descriptive modifiers to better determine which aspects of the condition contribute most to its being viewed with high or low regard (e.g., to aid in curriculum intervention planning).
We, therefore, sought to develop an attitude instrument for use with any medical condition, a scale on which respondents would rate the items “regarding patients with (the condition of interest).” This instrument is intended to capture a construct we are calling “medical condition regard,” which reflects biases, emotions, and expectations a given medical condition descriptor generates in the medical students and caregivers.
METHOD AND RESULTS
Our goal was to construct a scale for use across a wide range of conditions that would capture positive or negative biases, emotions, and expectations produced by medical condition descriptors. We first generated items consistent with the literature describing the responses of physicians to patients they like11 and dislike,12,13 and items reflecting discrediting responses seen toward stigmatized diagnoses.14 Second, to ensure these items were relevant to physicians' daily clinical experiences, we asked several primary care faculty members at our institution to review this item set and suggest improvements or new items. Finally, we incorporated our colleagues' suggestions into the pilot scale and eliminated items that were ambiguous, redundant, or not applicable across diverse medical conditions. The scale used for initial testing consisted of 18 items, nine positively and nine negatively worded.
We used a six-point Likert scale: 1 = strongly disagree, 2 = disagree, 3 = not sure but probably disagree, 4 = not sure but probably agree, 5 = agree, and 6 = strongly agree.
The sample consisted of medical students at Loma Linda University School of Medicine. Some participated in more than one phase of the testing. The minimum interval between participation times was six months (see Figure 1).
Testing the Scale
We conducted four sets of analyses to evaluate the scale: (1) reliability and exploratory factor analysis, (2) confirmatory factor analysis, (3) test—retest reliability, and (4) two tests of the scale's construct validity.
Reliability and exploratory factor analyses—method
The total participant pool consisted of 440 medical students (134 first-year, 119 second-year, 115 third-year, and 72 fourth-year students; 262 men and 178 women). We conducted the analyses in two phases. In the first phase, we gave the 18-item pilot scale to third- and fourth-year students. Third-year students were tested during their July orientation to clerkships; fourth-year students during the second month of their fourth year. To evaluate the scale's factor structure across diverse conditions, each student was given one of eight conditions to rate. The eight conditions involved two straightforward medical conditions (heartburn and pneumococcal pneumonia), one complex medical condition (severe emphysema from smoking), four psychiatric conditions (dementia, depression and intermittent suicidal thoughts, intravenous drug use, and long-standing auditory hallucinations and paranoid delusions), and one somatoform condition (long-standing abdominal complaints and no abnormality found on repeated evaluations). We expected these conditions to generate a good range of responses.6,7,11 Conditions were randomly assigned, and the numbers of students rating the individual conditions were roughly comparable.
Because the study of attitudes toward patients with addictive, somatoform, or other psychiatric conditions is a likely use of this scale, in the second phase of the exploratory factor analysis we evaluated the scale's factor structure using five such conditions and one comparison straightforward medical condition. In this phase, first- and second-year students (tested at the beginning of the school year) each rated one of six conditions: (1) acute meningitis; (2) alcohol dependence; (3) depression and intermittent suicidal thoughts; (4) longstanding auditory hallucinations and paranoid delusions; (5) panic disorder; and (6) frequent visits, many different symptoms, and few physical findings. Again, the conditions were randomly assigned (comparable numbers in all groups).
We evaluated frequency distributions for items, item-total correlations, item homogeneity (coefficient alpha), and various factor solutions (principal axis factoring) in the third—fourth-year and first—second-year subgroups and in the combined group. To protect against mean differences confounding the factor solutions, we set item scores within each condition to zero (by subtracting individual item scores from mean scores within each condition) before factoring.
Reliability and exploratory factor analyses—results
After evaluating frequency distributions for items, item-total correlations, item homogeneity, and various factor solutions from the first administration (third- and fourth-year students, n = 187), the second administration (first- and second-year students, n = 253), and the combined group (n = 440), it appeared an 11-item unidimensional solution was best in all three instances. Alpha coefficients all were comparable, as were the factor solutions, so we report the results for the combined data. Coefficient alpha for the scale, which we named the Medical Condition Regard Scale (MCRS), was .87, and all factor loadings were greater than .40 (Table 1 shows the items and their factor loadings). The scale appears to tap the degree to which medical students find patients with the given medical condition enjoyable, treatable, and worthy of medical resources. (Copy available from the first author.)
Confirmatory factor analyses—method
Our exploratory analyses found a one-factor (unidimensional) solution to be best. Exploratory factor analyses, in part, let the data determine the best factor solution. In contrast, confirmatory factor analysis begins with a hypothesized model and evaluates its goodness-of-fit to the data. We used this type of analysis to evaluate how well the data fit a unidimensional model.
Participants were 163 third- and fourth-year medical students (77 third-year and 86 fourth-year students; 102 men and 61 women). The third-year students were tested on the first day of their psychiatry clerkship; the fourth-year students one month prior to graduation. Each student rated two psychiatric conditions, major depression and alcohol dependence.
We used confirmatory factor analysis (a standard statistical software package) to test the fits of our hypothesized unidimensional model to the depression and alcoholism data. To evaluate model fit, we used the conventional χ2 test15 and the comparative fit index (CFI).16 A nonsignificant (p >.05) χ2 is desirable and suggests the model is an adequate representation of the data. The CFI is an estimate of the proportion of sample information explained by the model, and it can range from 0 to 1 (1 = perfect fit). Probabilities are not associated with fit indices (their distributions are not known); values above .90, however, are considered adequate.
Confirmatory factor analysis also generates standardized path coefficients for each scale item. Like factor loadings in exploratory factor analyses, the standardized path coefficients reflect the strength of correspondence of each item to the construct (in this case, “medical condition regard”) and are best interpreted as correlation coefficients.
Confirmatory factor analyses—results
We ran two confirmatory factor analyses (first-order single-factor solutions), one for major depression and one for alcohol dependence. Following initial runs for each condition, we fixed significant item-to-item covariance parameters and re-ran relaxed models. Table 1 presents the final standardized path coefficients. For both models, all path coefficients were significant (p <.05). Moreover, both models resulted in excellent fits to the data: depression χ2(df = 41) = 50.05, p =.16, and depression CFI =.98; alcohol dependence χ2(df = 34) = 45.84, p =.08, and alcohol dependence CFI =.98.
The alcohol-dependence model required our fixing more item-to-item covariance parameters than in the depression model (resulting in fewer remaining degrees of freedom); however, the two final solutions were remarkably similar.
Participants were 93 second-year medical students who used the final scale to twice rate (17-day interval) severe emphysema from smoking. We evaluated test—retest reliability with correlation (rtt).
MCRS scores were stable over the 17-day time interval. The test—retest reliability coefficient (rtt) was .84.
We used two analyses to assess the scale's construct validity. First, we examined how the scale ranked the diverse conditions used in the exploratory factor analyses. Second, we examined changes in attitudes, as measured by the scale, across a six-week psychiatry clerkship.
For the first validity analysis (condition rankings), participants were the original 440 medical students. Using the data collected during the exploratory factor analyses, we calculated mean scores (using only the 11 items included in the final scale) for each condition and ranked the conditions from high (more positive) to low scores. We expected the students to rate the straight-forward, easily treated medical conditions most positively and the difficult-to-treat (somatoform) conditions most negatively.6,7 We expected mid-range scores for the psychiatric conditions and the one complex, self-induced medical condition (severe emphysema from smoking).
In the second validity analysis (clerkship effects), 77 third-year medical students (44 men and 33 women) participated, and each used the scale to rate two conditions, major depression and alcohol dependence, at the beginning and end of their six-week psychiatry clerkship. Each student was anonymously identified by a self-chosen password to allow matching of pre- and post-clerkship responses. Because psychiatric clerkship experiences improve medical students' attitudes toward psychiatric patients,17 we hypothesized our students' mean ratings of patients with major depression would be more positive at the end of the psychiatry clerkship than at the beginning. To test this hypothesis, we used a matched-samples t test; alpha =.05.
At the time of this research, approximately 40% of our third-year medical students spent half their psychiatry clerkship on an addiction treatment program, and 60% spent all six weeks at general psychiatry sites. We hypothesized that the mean ratings of patients with alcoholism would increase more for the students who had spent three weeks on the addiction treatment program (n = 28) than for those without such exposure (n = 45). We evaluated difference score means (time 2 minus time 1) with an independent-samples t test (alpha =.05).
Table 2 presents MCRS means and standard deviations by condition. As expected, mean scores were highest for the straightforward medical conditions (highest regard) and lowest for the somatoform conditions. Scores for the psychiatry conditions and severe emphysema from smoking were in the mid-range.
The students' mean MCRS scores for major depression at the end of their clerkship were significantly higher than those on the clerkship's first day (52.6 ± 7.0 versus 49.4 ± 7.3; t[df = 76] = 4.74, p =.000). Scale scores increased (higher regard) for 55 of the 77 students, decreased for 20, and remained the same for two.
The MCRS detected a difference in attitude changes between the students who had rotated on an addiction treatment program and those who had not. The mean MCRS difference score (time 2 minus time 1) for students who had spent three weeks on the alcoholism unit was 4.39 ± 8.42 (range = +21 to −18), versus −.667 ± 8.52 (range = +18 to −20) for the students who had not done so. The five-point mean difference was statistically significant (t[df = 71] = 2.49, p =.016).
Medical Condition Regard Scale scores are reliable (coefficient alpha =.87, test—retest reliability =.84), and the scale appears to be a valid measure of regard for medical conditions. Although unidimensional, the final items (the same as in Table 1 reflect three themes: the degree to which respondents find patients with a given medical condition enjoyable, treatable, and worthy of medical resources. The enjoyableness and treatability themes resonate well with descriptions of physicians' reactions to patients they like and dislike.11,12,13 The medical-worthiness theme taps the discrediting or devaluing reactions that typify responses to stigmatized individuals.18
The MCRS performed as expected in our validity tests. Ranking the 12 study conditions by MCRS scores, the somatoform conditions scored lowest, consistent with studies finding high levels of somatization in patients whom physicians find frustrating or “difficult.”6 These patients generate feelings of exasperation, helplessness, and despair,12,13 feelings unlikely to be encountered in the care of patients who have the straightforward and more easily treated conditions with the highest MCRS scores. The MCRS also performed well in detecting expected attitude shifts over the course of a six-week psychiatry clerkship. Mean MCRS scores for patients with major depression rose significantly, whereas mean scores for patients with alcoholism rose only in the group of students rotating on an addiction-treatment unit.
The degree of variability seen in the change in MCRS scores across the psychiatry clerkship is intriguing. For example, although the mean MCRS score for alcoholism rose by four points for the students rotating on the addiction treatment program, changes in the scores of individual students (all of whom rotated on the same addiction treatment program) varied from +21 to - 18. This variability in magnitude and direction of change likely reflects both variability in clinical experiences and intra-individual factors. The durability of such score changes, and the factors contributing to them, deserve further study.
Although the MCRS appears to be psychometrically sound, and has, we believe, good potential for studying attitudes in medical students, physicians, and other medical caregivers, the data presented here are limited. We have no data as yet correlating MCRS scores with actual behaviors in clinical situations. Similarly, although relatively small changes in mean MCRS scores achieved statistical significance, it is not clear whether these represent changes of true educational or clinical importance.
Moreover, the students gave the lowest-scoring condition a mean MCRS rating of 37.4, which is above the scale's midpoint. This raises the possibility that the scale is vulnerable to ceiling effects. The relatively high scores could be due to our study sample's being weighted toward first- and second-year medical students, that is, students with limited exposure to the types of patients clinicians typically find difficult or frustrating. MCRS ratings of somatoform conditions by experienced clinicians, or clinicians whose career choices reflect an avoidance of such patients, might be substantially lower than those given by our students.
Finally, what this unidimensional scale captures, which we have called “regard,” requires further study. Conceptual questions include the relationship of regard to other constructs such as cynicism19 and tolerance of ambiguity,20 and the degree to which regard reflects respondents' personality characteristics. Practical questions include the responsiveness of regard to specific interventions and nonspecific factors (e.g., fatigue), and the durability of any changes produced.
In conclusion, we developed a non—condition-specific attitude scale for use with medical students and other caregivers designed to capture positive or negative biases, emotions, and expectations generated by medical condition descriptors. The scale demonstrated good psychometric properties and performed as hypothesized in tests of its validity. Although it needs further study, we believe this scale will be useful for investigations of attitudes toward medical conditions in educational and clinical settings.
1. Gerbert B, Maguire BT, Bleecker T, Coates TJ, McPhee SJ. Primary care physicians and AIDS: attitudinal and structural barriers to care. JAMA. 1991;226:2837–42.
2. Yedidia MJ, Berry CA, Barr JK. Changes in physicians' attitudes toward AIDS during residency training: a longitudinal study of medical school graduates. J Health Soc Behav. 1996;37:179–91.
3. Yedidia MJ, Berry CA. The impact of residency training on physicians' AIDS-related treatment practices: a longitudinal panel study. Acad Med. 1999;74:532–8.
4. Kopacz DR, Grossman LS, Klamen DL. Medical students and AIDS: knowledge, attitudes and implications for education. Health Educ Res. 1999;14:1–6.
5. Dickstein LJ, Hinz LD. The stigma of mental illness for medical students and residents. In: Fink J, Tasman A (eds). Stigma and Mental Illness. Washington, DC: American Psychiatric Press, 1992.
6. Lin EHB, Katon W, Von Korff M, et al. Frustrating patients: physician and patient perspectives among distressed high users of medical services. J Gen Intern Med. 1991;6:241–6.
7. Hahn SR, Kroenke K, Spitzer RL, et al. The difficult patient: prevalence, psychopathology, and functional impairment. J Gen Intern Med. 1996;11:1–8.
8. Hanna E. Attitudes toward problem drinkers, revisited: patient—therapist factors contributing to the differential treatment of patients with alcohol problems. Alcohol Clin Exp Res. 1991;15:927–31.
9. Ingersoll GM, Hibbard RA, Kronz KK, Fineberg NS, Marrero DG, Golden MP. Pediatrics residents' attitudes about insulin-dependent diabetes mellitus and children with diabetes. Acad Med. 1990;65:643–5.
10. Selfe SA, Van Vugt M, Stones RW. Chronic gynaecological pain: an exploration of medical attitudes. Pain. 1998;77:215–25.
11. Hall JA, Epstein AM, De Ciantis ML, McNeil BJ. Physicians' liking for their patients: more evidence for the role of affect in medical care. Health Psychol. 1993;12:140–6.
12. O'Dowd TC. Five years of heartsink patients in general practice. BMJ. 1988;297:528–30.
13. Gerrard TJ, Riddell JD. Difficult patients: black holes and secrets. BMJ. 1988;297:530–2.
14. Link BG, Cullen FT, Mirotznick J, Struening E. The consequences of stigma for persons with mental illness: evidence from the social sciences. In: Fink PJ, Tasman A (eds). Stigma and Mental Illness. Washington, DC: American Psychiatric Press, 1992.
15. Bollen KA. Structural Equations with Latent Variables. New York: Wiley, 1989.
16. Bentler PM. Comparative fit indices in structural models. Psychol Bull. 1990;107:238–46.
17. Singh SP, Baxter H, Standen P, Duggan C. Changing the attitudes of “tomorrow's doctors” towards mental illness and psychiatry: a comparison of two teaching methods. Med Educ. 1998;32:115–20.
18. Dovidio JF, Major B, Crocker J. Stigma: introduction and overview. In: Heatherton TF, Kleck RE, Hebl MR, Hull JG (eds). The Social Psychology of Stigma. New York: Guilford, 2000.
19. Testerman JK, Morton KR, Loo LK, Worthy JS, Lamberton HH. The natural history of cynicism in physicians. Acad Med. 1996; 71 (10 suppl):S43–S45.
20. Budner S. Intolerance of ambiguity as a personality variable. J Pers. 1962;30:29–50.