Conceptual model of predisposing, need, and contextual factors affecting patients' requests and physician behavior.
Definition of a request (reprinted from Kravitz et al21).
Kravitz RL, Bell RA, Azari R, Kelly-Reif S, Krupat E, Thom DH. Direct Observation of Requests for Clinical Services in Office PracticeWhat Do Patients Want and Do They Get It?. Arch Intern Med. 2003;163(14):1673-1681. doi:10.1001/archinte.163.14.1673
Copyright 2003 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2003
Requests can influence the conduct and content of the medical visit. However, little is known about the nature, frequency, and impact of such requests. We performed this study to ascertain the prevalence, antecedents, and consequences of patients' requests for clinical services in ambulatory practice.
This observational study combined patient and physician surveys with audiotaping of 559 visits to 45 physicians in 2 health care systems between January and November 1999. All patients had a new problem or significant health concern. Main outcome measures included prevalence of 8 categories of requests for physician action; odds of patients' requesting tests, referrals, or new prescriptions; odds of physicians' ordering diagnostic tests, making specialty referrals, or writing new prescriptions; patient satisfaction; and physicians' perceptions of the visit.
The 559 patients made 545 audiocoded requests for physician action; 23% requested at least 1 diagnostic test, specialty referral, or new prescription medication. Requests for diagnostic tests were more common among new patients (P<.001). Requests for any clinical service were more common among patients experiencing greater health-related distress (P<.05) and less common among patients of cardiologists (P<.001). After adjusting for predisposing, need, and contextual factors, referral requests were associated with higher odds of receiving specialty referrals (adjusted odds ratio [AOR], 4.1; 95% confidence interval [CI], 1.6-10.7) and prescription requests were associated with higher odds of receiving new prescription medications (AOR, 2.8; 95% CI, 1.2-6.3). Physicians reported that visits during which patients requested diagnostic tests were more demanding than visits in which no such requests were made (P = .02).
Though more common in primary care than in cardiology, patients' requests for clinical services are both pervasive and influential. The results support placing greater emphasis on understanding and addressing the patient's role in determining health care utilization.
CONSUMER EMPOWERMENT in health care is increasingly accepted as inevitable, right, and potentially beneficial.1- 5 However, with the more rigid forms of managed care in retreat and health insurance premiums rising,6 the need to control health care costs has never been greater. Physicians are frequently the target for cost control measures. Meanwhile, the role of the patient has undergone relatively little scrutiny. Requests are an important means by which patients can influence the conduct and content of the medical visit.7,8 To begin to understand the patient's contribution to health care resource use and quality, it is important to understand the epidemiology of patients' requests and their effects on physician behavior.
The behavioral utilization model formulated by Andersen and Newman9 views health services utilization as a function of predisposing, enabling, and need factors. A modified version of this model can be applied to patients' requests for services and their impact on physicians' clinical decisions (Figure 1). Predisposing factors include elements that might predispose patients to make requests or encourage physicians to fulfill them. Need factors include objective measures of health status as well as subjective evaluations of the need for health care. Enabling factors include contextual elements that might facilitate request behavior (Figure 1).
Previous studies have used surveys to assess patients' desires and to evaluate the match between patient and physician agendas.10- 14 However, survey data are subject to reporting bias. Direct observation of office visits using audiotape eliminates this bias and facilitates analysis of patient and physician behavior.
We tape-recorded a probability sample of ambulatory visits to 45 physicians practicing internal medicine, family medicine, and cardiology in northern California. We focused on patients with a new problem or significant health concern, screening out those attending for health maintenance or routine follow-up of a chronic condition. In so doing, we sought to address the following research questions:
What is the prevalence of patients' requests for 8 categories of physician action, and which request types are most common?
Which predisposing, enabling, and contextual factors are associated with making requests for 3 specific clinical services (diagnostic tests, specialty referrals, and new prescriptions)?
To what degree do patients' requests for tests, referrals, and medications influence physicians' provision of these services?
We suspected that patient requests for clinical services are both more prevalent and more influential than generally acknowledged by health economists and health care administrators.
The study was conducted within 2 health systems in the Sacramento, Calif, metropolitan area. This region has a population of 1.5 million that demographically resembles the United States as a whole.15 Data were gathered from patient and physician surveys and from audiotape recordings of problem-driven visits. Given our focus on sick care, well-visits for preventive care, and routine follow-up of chronic problems were excluded.
Two health care systems representing the dominant systems of care in the region were selected as research sites: the University of California Davis Medical Group (UCDMG) and Kaiser-Permanente (Kaiser). The UCDMG is a large multispecialty group practice with explicit utilization review where physicians are paid fee-for-service according to a relative value scale. Kaiser is a group model health maintenance organization where physicians are paid on salary. Human subjects approval was obtained from institutional review boards at both UC Davis and Kaiser.
Clinicians practicing family medicine, general internal medicine, or cardiology at least 20 hours per week were eligible to participate. A total of 45 physicians volunteered for the study, 22 from UCDMG and 23 from Kaiser. Sixteen practiced general or family medicine, 18 general internal medicine, and 11 cardiology. The mean age of participating physicians was 44 years and 31% were women. Because physician recruitment relied on the personal efforts of the investigators and their staff, overt refusals to participate were rare. Participating physicians were similar to all Kaiser and UCDMG primary care physicians in terms of age (mean, 44 years) and sex (25% female).
Owing to the length of the patient survey and complexity of consent procedures, patients were recruited by telephone in advance of a scheduled visit rather than in the clinic waiting area. Enrollment occurred by practice in waves between January and November 1999. Each weekday evening during the study period, a research assistant reviewed next-day appointment lists obtained from participating physicians (Monday's lists were reviewed on Friday). A research assistant contacted patients in random number order by telephone until quotas (approximately 20 patients per physician) were met. A total of 4560 patients were selected for telephone contact, and after 3 telephone calls per household, 2606 patient telephone contacts were made (contact rate, 57%). Patients were eligible for study if they spoke English, could complete a written questionnaire, and (1) had a new or worsening problem (answered yes/no) or (2) were at least "somewhat concerned" (≥3 on a 1-5 scale) that they might have a serious, undiagnosed disease. We imposed this requirement because problem-driven visits account for about two thirds of visits to general internists and offer richer opportunities for clinical negotiation and decision making.16 Among study entrants, 26% were eligible on the basis of having a "new problem," 19% because they were at least "somewhat concerned" about a serious diagnosis, and 55% because they met both criteria. Among those with health concerns, 60% were somewhat concerned, 26% very concerned, and 14% extremely concerned.
Among the 4560 patients in the original sample, 909 provided usable study data, 423 were eligible but not enrolled, 2407 were of unknown eligibility, and 821 were confirmed ineligible. The net corrected response rate (usable responses divided by estimated number of eligibles) was 909/2824 = 32.2% and the cooperation rate (survey completion rate among those known to be eligible) was 909/1332 = 68.2%.17 Approximately 95% of visits (860/909) were successfully audiotaped. Because of limited study resources, 560 tapes were selected by simple random sampling for detailed interaction analysis (1 tape had inaudible segments and was eliminated, leaving 559). Power analysis indicated that 540 patient visits were required to detect with 80% power a 10% increase in the proportion of patients requesting or receiving clinical services (eg, comparing males with females), assuming a baseline request (or receipt) rate of 15%. All patients gave written informed consent and received a $10 honorarium for completing the study.
Survey data were obtained from patients and physicians. Patients reported on age, sex, ethnicity, and years of education. Health-related distress was assessed as a 3-item composite of the extent to which patients were bothered by their symptoms, worried about their health, and concerned that they might have a serious illness (each item measured from 1 [not at all] to 5 [extremely]) (α = .70). Previsit trust in the treating physician was measured using a previously validated scale,7,18 and postvisit satisfaction was measured as the mean of 4-items rated on a 5-point scale (ranging from excellent to poor).19 Physicians indicated the type of visit (comprehensive evaluation of new patient, follow-up of known patient, urgent visit for patient of colleague, other); checked a list of 14 possible comorbid conditions (mean, 1.46; range, 0-7); and reported whether, during the visit, they ordered or performed a laboratory or imaging test, made a referral to a physician specialist, or prescribed any new medications. They also made ratings on two 5-point scales of (1) satisfaction derived from the visit (1 = far less satisfying than the typical visit, 5 = far more satisfying) and (2) the demands of the visit (1 = far less demanding than the typical visit, 5 = far more demanding).
Patient requests for services were coded using the Taxonomy of Requests by Patients, a valid and reliable request classification scheme previously described.20- 22 Guided by a standardized manual (available from R.L.K.), 2 trained assistants (coder A and coder B) listened to the tapes until they came to a "question, command, statement or conjecture that a native speaker would recognize as a desire for information or action" (Figure 2). They classified all such utterances into 1 of 12 categories of requests for information (not reported here) or 8 categories of requests for action (diagnostic tests, new medications, specialty referrals, physical examination, medication refills, therapeutic procedures, nonphysician referrals, and "other") (Table 1). When raters concluded that a patient statement or question might represent a subtle, inexplicit request for physician action, they coded the question or remark as an information request but tagged it as a "veiled request for action" (Table 2). To maintain quality and assess reliability of the coding process, a random sample of 95 tapes were double-coded by a supervisor. As in previous studies,20,21 overall unitizing reliability (indicating consistency in identifying requests) was moderate (κ = 0.64) and classification reliability (indicating consistency in sorting requests into the 20 Taxonomy of Requests by Patients categories) was high (mean κ = 0.75 for information requests and κ = 0.82 for action requests).
Analyses were corrected for the clustering of patients within physicians using the Stata 6.0 svymean, svytab, and svylogit procedures for complex surveys.23 In these analyses, the treating physician was identified as the primary sampling unit and a 6-level stratification variable was created by crossing specialization (internal medicine, family practice, and cardiology) with site (UCDMG, Kaiser). As in other studies where interest centers on the population of patients in a practice (rather than just those who show up for care),24 probability weights were assigned to the patients to account for differences in volume and participation rates among practices. Thus, the weighted results reflect estimates of probabilities that would have been observed had the entire population of patients in these practices taken part in the study.
Patient requests were analyzed at both the request level (without adjustment for design effects) and at the patient visit level (with adjustment). Differences in the proportion of patients requesting tests, referrals, or medications according to personal and contextual characteristics were assessed using the svymean and svytest procedures in Stata 6.0 (Table 3). The effects of selected predisposing, need, and contextual characteristics (Table 4) on the odds of making at least 1 request for tests, referrals, or new prescription medications were evaluated in separate design-corrected logistic regressions using the svylogit procedure. Using a similar set of covariates (Table 5), we also assessed the influence of patient requests on the odds of receiving a test, referral, or new prescription, again using svylogit in Stata 6.0. Finally, we used multiple linear regression to examine the effect of patients' requests for services on physician satisfaction and perceptions of visit demands after controlling for the covariates in Table 5 as well as visit length. All reported P values are 2-sided.
Patients selected randomly for audiotape analysis in this study (n = 559) were similar to patients without audiocoding data (n = 350) in terms of age, sex, ethnicity, education, health-related distress, and physician specialty, but they were more likely to have 2 or more chronic conditions (46% vs 37%) and to receive care from the staff model health maintenance organization (56% vs 44%). In subsequent analyses, site of care was not significantly associated with any outcomes of interest. Patients in the analytic sample had a mean age of 55 years; 57% were women, 19% were nonwhite, 31% had graduated from college, and 78% had at least 1 physician-reported chronic condition. Seventy-three percent were making follow-up visits with their own physicians. One third of visits were to female physicians and about one fourth were to cardiologists.
The 559 patients made a total of 545 audiocoded requests for physician action. Nearly one fourth of patients made at least 1 explicit request for all or part of the physical examination, 11% for 1 or more new prescription medications, 9% for a medication refill, 8% for diagnostic testing, and 5% for a physician specialty referral (Table 1). Twenty-three percent of patients requested at least 1 test, new prescription, or referral. Most patients made between 0 and 2 requests per category (data not shown). Among patients making at least 1 request of a particular type, the mean number of requests per patient ranged from 1.0 (referral to a nonphysician, eg, physical therapist, podiatrist, or optometrist) to 1.67 (physical examination) (Table 1). In addition to these overt requests, there were an additional 74 "veiled requests for action" (Table 2). As judged on audiotape, 13% of patients made at least 1 action request that was overtly denied, passively skirted, or incompletely fulfilled (data not shown).
Table 3 shows the weighted percent of respondents making requests for at least 1 test, referral, new prescription medication, or for any of the 3 services, arrayed according to predisposing characteristics (patient age, sex, education level), clinical need (comorbidity count, healthrelated distress), and contextual factors (visit type, physician specialty, system of care). Requests for any service were much more common among patients of internists (27%) and family physicians (22%) than among patients of cardiologists (8%). Women were more likely than men to make requests for specialty referrals and for "any service" (Table 3). Patients with greater health-related distress were more likely to request any service. Request behavior was unrelated to age, level of education, or the number of chronic conditions. New patients undergoing comprehensive evaluation were more likely to request diagnostic tests, whereas follow-up patients were more likely to request new prescription medications (Table 3).
Using design-corrected multivariable logistic regression, we assessed the independent effects of predisposing, need, and contextual factors on patients' requests for services. Requests for testing were significantly more common during new visits (adjusted odds ratio [AOR], 3.2; 95% confidence interval [CI], 1.4-7.2; Table 4). Requests for referral and for new prescriptions were less common in cardiology (for referrals: AOR, 0.10; 95% CI, 0.01-0.85; for new prescriptions: AOR, 0.35; 95% CI, 0.14-0.93). The odds of making any request for services (tests, referrals, or new medications) were significantly greater among patients with more health-related distress (AOR, 1.3 per level of increasing distress; 95% CI, 1.0-1.7) and less among patients of cardiologists (AOR, 0.29; 95% CI, 0.14-0.60) (Table 4). In ancillary regressions, terms for patient trust in the treating physician, for visit length (in minutes), and for physician age and sex did not materially alter the results and, with one exception, were not statistically significant. The sole exception was that patients visiting female physicians were significantly more likely to request a specialty referral (AOR, 2.23; P = .045).
After adjustment for predisposing, need, and contextual factors, we evaluated the effects of making requests for diagnostic tests, specialty referrals, or new prescriptions on receipt of these respective services. The odds of receiving a specialty referral were significantly increased among patients making a request for referral (AOR, 4.1; 95% CI, 1.6-10.7). The odds of receiving a new prescription were similarly increased among patients making a request for medication (AOR, 2.8; 95% CI, 1.2-6.3; Table 4).
Looking at other components of the model proposed in Figure 1, testing was significantly increased among patients with more chronic conditions and those making new comprehensive visits. Referrals were more common among patients with greater health-related distress (P = .055). Referrals were less common in cardiology (P<.05). New prescriptions were less common with advancing age (P<.001, Table 5), although the number of refilled prescriptions increased with age (r = .18, P = .0001, data not shown in tabular form).
Patients whose physician failed to fulfill 1 or more action requests (n = 41) registered significantly less visit satisfaction (mean satisfaction scale score = 4.06, consistent with an overall rating of "very good") than patients whose every action request was fulfilled (n = 261, mean satisfaction score = 4.37) and patients who made no action requests (n = 257, mean satisfaction score = 4.38). As in previous work based on self-report,7 there was no difference in satisfaction between patients who had all action requests fulfilled and those who made no such requests.
After adjustment using multiple regression, physicians' ratings of their satisfaction with the visit were unaffected by patients' requests for services. However, physicians' perceptions of visit demands were significantly increased among patients who requested diagnostic tests (averaging 0.31 points higher on a 5-point scale; P = .02). Perceptions of visit demands were not increased among patients who requested referrals or new prescriptions.
In this study of problem-driven visits, patients made direct, codable requests for clinical resources in about one fourth of primary care encounters but only 8% of cardiology visits. The differences persisted after adjusting for various factors that might on a theoretical basis be expected to influence request behavior. The observation that request density is not uniform across specialties suggests that the consumer movement has greater traction within some quarters of the health care system than others.
What characteristics of primary care vs specialty practice might explain this? Perhaps most important, the scope of practice in primary care is broader. Primary care physicians typically evaluate more conditions per visit than do cardiologists. More conditions provide more opportunities for patient requests. In addition, procedurally oriented subspecialists often see patients for the explicit purpose of arranging diagnostic tests (such as angiography), so patients need not ask directly for investigation of their symptoms. Patients realize that primary care physicians and subspecialists have different professional roles and that making referrals for problems unrelated to their expertise is beyond the scope of subspecialty practice. Patients may also assume that subspecialists have greater intellectual mastery over the therapeutic options within their clinical domain25,26 and need not be asked for specific therapies, including medications.
The differences in prevalence of requests for services between specialty sectors are not explained by differences in encounter time or by enhanced familiarity with the treating physician. Furthermore, the differences are consistent with survey data showing that primary care physicians are more troubled than specialist consultants by direct-to-consumer advertising, which increases drug sales by provoking patient requests for prescriptions.27,28 Whatever the reasons, the results suggest a different dynamic within primary care and subspecialty practice and raise questions for future research about the applicability of shared decision making across clinical disciplines.29
These results have implications for primary care practice. With requests for services featuring in about 25% of primary care visits, explicit instruction in how to address them should be an integral part of primary care training and continuing education. This imperative is underscored by the observation that patients' requests (at least for diagnostic tests) are associated with increased physician stress (as measured by their perception of visit demands).
Although direct requests for tests, medications, and referrals were common, they were far from universal. There are several possible explanations. First, many patients are simply not inclined to play an active role in their care. In one recent study, up to 34% of women recently diagnosed as having breast cancer wanted to delegate all decision making to the physician.30 Second, patients tend to trust their physicians to make appropriate recommendations for health care resource use.31,32 Third, patients can engage in the process of clinical negotiation and influence the process of care without making direct requests. In a recent study by Scott et al,33 many patients desiring antibiotics for upper respiratory tract infections did not ask for them directly. They instead suggested a candidate diagnosis, described symptoms characteristic of a particular diagnosis, emphasized the severity of symptoms, appealed to life-world circumstances, and shared stories of previous antibiotic efficacy. Although, we attempted to account for such veiled requests, we were likely to have missed many of them.20,21 Our inability to fully enumerate indirect requests probably explains some of the 2- to 4-fold differences between the prevalence of audiocoded requests reported here and the prevalence of self-reported requests published previously.7 Additional research is needed to determine other sources of this gap, including differences in how patients, physicians, and researchers define requests.
Among the possible antecedents of patient requests for services, only health-related distress, visit type (being a new patient), and physician specialty were significant predictors. Being female was marginally significant (P = .08). Subjective health distress predicted requests for services more powerfully than did an objective count of chronic conditions. The latter is an important predictor of utilization34,35 but not requests, which may be driven more by anxiety than disease burden. New patients may simply have more unmet health care needs and expectations.14 Women are known to use more health care than men, to play a dominant role in making health care decisions for the family, and to interact more assertively than men in health care settings, all of which might lead them to make more requests.36- 38
Patient requests were a strong independent predictor of physician referral and prescribing. While we are confident of this conclusion in general terms, we do not know the degree to which specific requests resulted in provision of specific services. For example, a patient may have requested an antibiotic and received an antihistamine. Nevertheless, at least within broad categories, patient requests are powerful. Physician decision making has been conceptualized as a function of the clinical characteristics of patients, physician practice style, and economic incentives.39 Classic 2-part econometric models assume that patients decide when to seek care initially but providers determine subsequent use.40 Our results suggest that patients' influence is more pervasive, affecting care at different points along an episode of illness or relationship. An alternative explanation is that unmeasured case-mix (eg, severity of illness, comorbidity, and anxiety) simultaneously increases both service utilization and patient requests for services. However, the reported odds ratios were large and consistent with other studies showing the substantial impact of patient demand on physician prescribing of other services such as antibiotics.41 One implication is that programs for containing health care costs must address the patient as well as the physician. It is perhaps unrealistic to expect that physicians will refuse to provide clinical services to patients whose requests are sufficiently strident.42,43 Although the evidence from studies of patient decision aids is variable, more informed patients often select more conservative and less costly treatments.44 Therefore, efforts to make evidence-based medicine accessible to patients and physicians may be a promising strategy for utilization management as well as quality improvement.
Physicians experienced visits in which patients requested diagnostic tests as particularly demanding. The results are broadly consistent with a study initiated by Kravitz et al21 in the mid-1990s. In that study of 131 audiotaped visits, physician perceived greater cognitive and emotional demands when patients made more information or action requests. In the present study, the special salience of diagnostic test requests may derive in part from the study entry criteria: patients sufficiently concerned about a "new or worsening problem" or a "serious undiagnosed condition" to request a test may be particularly anxious and difficult to manage. They may also be more likely to pose a diagnostic dilemma.45
These results should be interpreted in light of several study limitations. The Taxonomy of Requests by Patients was not designed to capture nonverbal requests for services, and it has only limited capacity to identify verbally inexplicit or "veiled" requests, which may be very common. The study was limited to 45 physicians practicing 3 specialties in a limited geographic area. We focused on patients with current health concerns, excluding those who were making routine follow-up visits for chronic disease management as well as those seeking preventive health care. The request patterns of such patients might be different. We did not distinguish between requests that were highly significant to patients and those valued less highly. Nevertheless, we conclude that patient requests for services are common across diverse practice settings and appear to have an important relationship with physician behavior. Research focused on understanding the role of the patient in the clinical negotiation will support the development of new strategies for containing costs and improving appropriateness of care.
Corresponding author and reprints: Richard L. Kravitz, MD, MSPH, Center for Health Services Research in Primary Care, University of California, Davis, Patient Support Services Building, 4150 V St, Suite 2500, Sacramento, CA 95817 (e-mail: email@example.com).
Accepted for publication September 30, 2002.
This study was supported by grant 034384 from the Robert Wood Johnson Foundation, Princeton, NJ.
We are grateful to the patients and physicians at participating practices for their time and effort; Sara Lu Vorhes for project management and data collection; Lisa Silverio, Robert Dominguez (deceased), and Carrie Willis for audiotape coding; Christine Harlan for general project support; and Peter Franks, MD, for a helpful review of an earlier version of the manuscript.