Effect on Health-Related Outcomes of Interventions to Alter the Interaction Between Patients and Practitioners: A Systematic Review of Trials

Identification of Relevant Trials and Data Extraction

We developed the search strategy using Delphi consensus methods (SJG, MWMV, ALK, MS) (Figure 1⇓). Each term was combined with the Cochrane trials search strategy¹⁶ in MEDLINE (1966–1999), HealthSTAR (1975–1999) and PsycINFO (1967–1999) databases. We then hand-searched bibliographies of trials meeting the inclusion criteria. Titles and abstracts (where available) of identified reports were assessed. Full copies of reports resembling relevant trials were obtained. At least 2 authors independently checked them against inclusion criteria and extracted data onto standardized forms. Discrepancies were assessed by an independent researcher and resolved by discussion (SJG, MWMV, SG, JG, ALK, MS).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Search terms.

Inclusion Criteria

Studies set in primary or secondary medical or nursing care were included. Participants had to be allocated at random to an intervention aimed at altering the interaction between patients and practitioners within a consultation or office visit. The evaluation had to include assessment of a health outcome or satisfaction. Studies in any language were included.

Quality assessment took into account established criteria predicting the likely presence of bias.¹⁷ Criteria included allocation concealment at randomization and outcome assessment, losses to follow-up, and adjustment for cluster design, if present.

Development of Taxonomies of Intervention Type and Measures

Interventions used in trials to alter patient-practitioner interaction are varied and difficult to classify. We developed a taxonomy based on our theoretical understanding of the ways in which collaboration between patient and practitioner can be enhanced and may lead to improved outcomes. This framework has been described as the new clinical method.¹⁵ It involves the patient and practitioner reaching a mutual understanding of each others’ explanatory models of illness and disease. It therefore takes into account the patient’s ideas, concerns, and expectations at all stages of the clinical interview, from gathering information through problem formulation to decisions about management. It is hypothesized that this method will lead to more precise framing of the problem and management that makes sense to the patient, which could enhance outcomes through several mechanisms. Specifically, a correct diagnosis will be matched to a more effective prescription. Involvement of the patient will lead to engagement in taking any agreed treatment. It will also mobilize a range of self-care activities tending to increase well-being and function through social, psychological, and perhaps even immunological mechanisms.¹⁸ Finally, the collaborative approach might reduce the mismatch between a patient’s and the practitioner’s representations of the problem and thereby increase efficiency through a reduction in unnecessary investigations, consultations, and unused medication.¹⁹

Approaches to changing the interaction between patient and practitioner within this framework can be aimed directly at patients (for example by a preconsultation leaflet or interview), via the practitioner (for example through courses), or at both parties. Components of interventions that we consider central to the new clinical method include those aimed at (1) increasing the patient’s contribution at all stages of the clinical interview (patient activation), and (2) increasing the practitioner’s ability to elicit and integrate the patient’s views with the biomedical view (communication skills training).

More specifically, interventions can aim to (1) increase the patient’s understanding of the biomedical model (provision of information about disease or treatment; discussion of behavior change), and (2) increase the practitioner’s understanding of psychosocial aspects of the illness (attention to patient’s emotion).

Having specified these components, we refined them by reading the studies and developed brief descriptions of each study intervention and comparison approach. Three authors undertook this process independently prior to consensus discussion (ALK, SJG, MS, SG). This effort added 2 further intervention subtypes: one where the practitioner receives information by a note or questionnaire before the clinical encounter to alert him or her to a patient’s concerns or functional status (practitioner activation by patient questionnaire), and one where the practitioner changed the style of consultation between a biomedical or patient-centered approach (changes style). The extent to which these components were associated with improved health outcomes was then explored in the analyses.

A wide range of process measures was used. We defined measures of communication as subjective (self-report) or objective (videotape, for example). Process measures most often assessed among patients included frequency of question asking, controlling behaviors, and overall involvement in the consultation, particularly the decision making. They also included patients’ ratings of the extent to which the practitioner offered relevant, understandable information and explanation, and the extent to which patients felt understood and helped. Those processes measured among practitioners included the degree to which they elicited patient concerns, exhibited empathy, or considered psychosocial aspects of presenting complaints.

We classified health outcomes as objective measures of disease processes (for example, blood pressure and blood glucose concentrations), or subjective measures of illness experience by self-report (symptoms, anxiety, depression, functional status, well-being, and quality of life). We also included separately patient satisfaction with care and other health-related outcomes (health service costs, patient knowledge, and health-related behaviors including treatment adherence). The principal outcome for each study, when not specified by the author, was defined from sample size calculations, hypotheses, or aims.

Analysis

Descriptive analyses were undertaken of study settings, participants, quality, interventions, process, and outcome measures. Studies were too heterogeneous for meta-analysis. We grouped studies according to intervention type. Study process and outcome measures were reported as favoring intervention or comparison groups with accompanying probability (statistically significant at P <.05). We assessed the extent to which principal outcomes favored intervention or comparison groups across all studies. Then, for those with health outcomes, we assessed the effect of each prespecified category of intervention. The denominator for each comparison was the number of studies reporting the relevant outcome.

Search Strategy

More than 21,200 reports were identified, of which 148 were potentially relevant and fully assessed. Thirty-five trials met the inclusion criteria. The most common reasons for exclusion were nonrandomized design, no reported health outcomes, and the intervention not specifically targeting patient-practitioner interactions. Most were reported after 1995. Two similar but distinct trials were reported in one article.²⁰ Data were incomplete for one trial despite correspondence with the author.²¹ Characteristics of included trials are summarized in Table 1⇓.

Participants and Settings

Most trials were undertaken in primary care in North America. All were reported in English. More than 13,500 patients (median 200, interquartile range, 105–380) participated. Most trials involved patients who had a chronic problem (n = 15 trials), including cardiovascular disease, diabetes, epilepsy, depression, or cancer of breast or prostate, or they enrolled a sample of patients consulting in primary care (n = 11).

Trial Quality

Trial quality varied. Descriptions of allocation concealment at randomization and outcome assessment were frequently inadequate or missing (21 and 18 trials, respectively). Follow-up was short (median 4 weeks, interquartile range 0–26). Losses to follow-up were either not reported or exceeded 30% in more than a quarter of studies. Clustering was present in 15 of the included trials^{13,31,35–37,39–45,47–},⁴⁹; in 10 some adjustment was made for this clustering in the analysis.^{13,36,39,41–45,48,49} Health-related principal outcomes were measured objectively in 6 studies and subjectively in 19 studies. Four measured symptom resolution,^29,32–,³⁴ 10 measured anxiety or depression,* and 5 measured well-being or functional status.^{22–24,47,51} Two studies measured adherence to treatment.^43,48 In 8 studies satisfaction was the principal outcome. Only 4 studies with health outcomes met our pre specified quality criteria.^33,42,51,52

Characteristics of interventions are summarized in Table 2⇓. Eleven interventions were aimed at the patient directly, 16 were administered by the practitioner, and 8 used a combination of these approaches. Most interventions (n = 23) involved activation of the patient in the consultation, which included stimulating patients to take a more active role, for example, by considering questions to ask in advance of the consultation (n = 18),^† or writing down needs and expectations to be given to practitioners, thus activating them in turn (n = 5).^47–,⁵¹ Eleven studies evaluated training practitioners in communication skills.^‡ Skills included listening, eliciting patients’ views, and negotiating treatment plans.³⁰

In 12 studies^§ information about disease or treatment and in 7 discussion of behavior change^{30,36,37,39,40,41,46} was a main feature of the intervention. In 14 studies there was explicit attention to emotion (for example reference to affect, trust or patients’ concerns),^|| always in the context of more complex interventions. Classification of attention to emotional issues was the hardest on which to agree. In 4 studies practitioners simply altered their consultation style.^29,32–,³⁴

Interventions most commonly incorporated booklets, letters, or tapes either alone or in combination with training sessions, and they varied widely in intensity and complexity. Comparison groups were split fairly evenly between usual care (n = 18)^¶ and attention control (n = 17).^#

Effects on the Process of the Consultation

Effects of the intervention on the process of the consultation were measured in most trials (30 of 35, 86%), usually by patient (n = 21) or practitioner self-report (n = 11). Objective measures were used in 15 studies (patient, 10; practitioner, 12).** Processes significantly favored the intervention groups in most studies (22 of 30, 73%). In 6 studies at least 1 process deteriorated^{20,34,44,45,49,52} (2 significantly^34,49), although in 4 of these studies other process measures significantly favored the intervention.^20,44,45,49 (The effects of the interventions on the effects of consultations are displayed as supplemental data in Table 2a⇑, which is available online only at: http://www.annfammed.org/cgi/content/full/2/6/595/DC1).

Effects on Patient Outcomes

Table 3⇓ summarizes the effects of the interventions on outcomes and the direction of results in relation to the main intervention categories.

Trial Quality

This review of trials of interventions to alter the interaction between patient and practitioner and their effects on health-related outcomes is the most comprehensive yet undertaken. It shows clear weaknesses in the field. Only 35 trials up to the end of 1999 were identified by clear criteria. Studies tended to be small and short, with major design limitations. Only 4 studies with health outcomes met our pre specified quality criteria, which were selected for their ability to predict the likely presence of bias.¹⁷ It could be argued that these criteria could be wider still; for example, populations were often poorly characterized, and measures of process and outcome were unvalidated. Other reviews have analyzed these shortcomings in detail.¹² Here we focus on the limitations in description and evaluation of the interventions, which were often multifaceted, poorly described, and not clearly linked to measures of their likely effects. The mechanisms by which components of the intervention were hypothesized to affect processes within the consultation to influence outcomes (the underlying causal models) were rarely made explicit, and only once tested.⁵³ Such testing is essential for the field to advance. For example, this review showed the difficulty of isolating the effects of interventions to promote question asking. Kidd et al⁵⁴ have recently drawn on a theoretical framework from psychology to overcome cognitive and affective barriers to question asking. Brief interventions to promote question asking are based on the studies by Roter,⁵⁵ Greenfield et al,²² and Thompson.²⁰ In a randomized design, interventions were offered to patients attending a hospital diabetic clinic. Audiotapes were used for objective assessment of question asking. Those in the intervention groups showed an increase in self-efficacy of question asking after the consultation, and less anxious patients asked more questions. No significant differences were found between intervention, attention-control, and pure control groups in the number of questions asked, however. All groups asked more questions than in previous studies, suggesting a possible ceiling effect. This example shows forcibly the need for stronger research before implementing apparently simple, feasible, but possibly ineffective, interventions in practice.

Taxonomy of Process and Outcome

The taxonomy clarifies the wide range of measures of process and outcome used. Measures may be self-report or objective, both are clinically important and both have predictive validity.^10,56–,⁵⁹ We differentiated health outcomes into self-reports of the illness experience and objective assessment of the disease process. Practitioners have traditionally placed more emphasis on disease processes, but patients are naturally concerned also with current well-being. Showing that positive effects on illness experiences, such as well-being, can coincide with negative effects on disease processes, such as dyslipidemia,³⁹ emphasize further the need to measure both. Disease processes, however, were rarely measured.

Satisfaction, while an important health-related outcome, was excluded from our analyses of health outcomes because measures frequently overlap with the patients’ subjective assessment of the consultation process.^60,61

Impact on Process and Outcomes

The weight of evidence reviewed supports the hypothesis that the way practitioners and patients interact in the clinical encounter can be measured and altered. This finding is supported by other reviews.¹² The effect on health-related outcomes is more equivocal. If weight is given to the direction of findings, principal outcomes favored the interventions in almost three quarters of the studies; however, results reached statistical significance in less than one half of the studies. Among the 4 trials meeting our quality criteria, only 2 interventions were associated with significant improvements in the principal outcome. Only 2 of 6 studies with disease processes as the principal outcome significantly favored the intervention group, and 1 of 6 significantly favored the control group. Analysis by intervention type provides some promising directions, although study numbers become small.

Analysis by Intervention Type

Interventions could affect health outcomes whether aimed at patient, practitioner, or both. Almost one half of those interventions delivered via practitioners alone were associated with at least one worse outcome than the comparison group, whereas none was when delivered directly to patients. This finding is compatible with a greater positive effect of interventions directed to patients, and perhaps with the difficulties of consistently altering practitioner behavior.⁴⁰

Interventions including components directed at activation of patients, activation of practitioners by preconsultation note or questionnaire from patient, and provision of information and attention to emotion, all showed promise. significantly positive outcomes favoring the intervention group were found in more than one half of these studies. Patient activation was the approach most frequently evaluated, and there were some discordant findings.^23,24,40 Two complex interventions involving patient activation among groups with chronic disease, evaluated in the United States,^14,21 reported particularly large benefits, not achieved in pragmatic trials in the United Kingdom.^39,40 This work merits replication.

Complexity was not necessary for success. Patients who simply provided practitioners with written information about their needs, emotional concerns, and functional status in advance of the consultation were significantly less anxious, or showed improvement in functional status afterward. Such written information may work as a substitute for the failure of doctors to elicit patients’ ideas, concerns, and expectations, a failure that is well documented.⁷ Review findings also emphasize the centrality of appropriate provision of information to effective consulting^1,39 and the need to further develop and evaluate affective elements. We found insufficient evidence to support a particular style of consultation in general practice or to resolve the question of whether interventions to increase patient participation prolong consultations. Further work is needed.

Review Limitations

The review has a number of strengths: we had clear inclusion criteria, a comprehensive search strategy for published trials, assessment of trial quality, attention to a priori classification of intervention types, and careful accounting of process and outcome measures. Presentation of direction, as well as statistical significance of findings, allows consideration of the overall pattern of results. The review also has limitations. The search strategy was restricted to 3 databases and to the end of 1999. Relevant studies may not have been indexed with our search terms or may be unpublished. A recent review with a broader search strategy, however, did not include any trials that met our inclusion criteria but were omitted from our review.⁹ Moreover, a related review including the year 2000 identified only 3 further trials with health-related outcomes. These trials did not challenge our findings.¹² In addition, unpublished studies are more likely to have negative findings,⁶² which would only strengthen our conclusion that the evidence base is weak.

After initial unsuccessful attempts at correspondence, data were extracted and interpreted without reference to original authors, which may have led to some underestimation of trial quality and miscoding of intervention categories.⁶³ Interventions were multifaceted and consequently were classified into more than one category. Conclusions about effectiveness of individual components of interventions are therefore limited, and none can be drawn concerning the relative effects by patient age, sex, disease and ethnic group. The heterogeneity of populations, interventions, and measures precluded overall meta-analysis, although the possibility of further synthesis among subgroups of interventions, for example, to encourage question asking by patients, or improve practitioners’ communication skills, should be considered.

Implications and Conclusions

Successful interactions between patients and their practitioners lie at the heart of medicine, yet there are few rigorous trials of well- specified interventions to inform best practice. Trial evidence suggests that a range of approaches can achieve changes in this interaction and some show promise in improving patients’ health. In terms of practice there are strong justifications unrelated to evidence-based medicine for adopting a collaborative approach to the medical encounter, such as, for example, patient preferences and moral imperatives.¹⁵ This review demonstrates the potential for the way patients and practitioners behave in consultations to have measurable effects, both positive and negative, on their future health. In terms of research, it demands collection of better evidence to inform practice. Improvements are needed across the board, from characterization of study populations to study designs. Outcome measures themselves should include both objective measures of disease process and self-report of illness experience where appropriate. Interventions should be more carefully developed and specified. Rigorous explanatory trials of such interventions are still needed before pragmatic studies of cost-effectiveness. To advance knowledge further, we need to replicate promising studies using rigorous methods. These studies should include explicit theoretical frameworks for interventions, designed to link effects on key characteristics of the interaction through to effects on health outcomes. Only in this way will the field advance beyond a series of tantalizing but disconnected and unconfirmed results.