Do physicians’ recommendations pull patients away from their preferred treatment options?

Abstract

Context and objective  Shared decision making is especially advocated for preference‐sensitive decisions. We investigated whether physicians’ recommendations pull patients away from their preferred treatment option when making a preference‐sensitive decision.

Design, participants and methods  Inpatients (N = 102 with schizophrenia, N = 101 with multiple sclerosis) were presented with a hypothetical scenario (the choice between two drugs). They were first asked about their preferences concerning the two drugs and then they received a (fictitious) clinician’s recommendation that was contrary to their preferences. Subsequently they made a final choice between the two drugs.

Main outcome measures  The main outcome measure was whether the patient followed the physician’s advice in the hypothetical scenario. Thereby patient’s (pre‐recommendation) preferences served as a baseline.

Results  In the decision scenario, about 48% of the patients with schizophrenia and 26% of the patients with multiple sclerosis followed the advice of their physician and thus chose the treatment option that went against their initial preferences. Patients who followed their physician’s advice were less satisfied with their decision than patients not following their physician’s advice (schizophrenia: t = 2.61, P = 0.01; multiple sclerosis: t = 2.67, P = 0.009).

Discussion and conclusions  When sharing decisions with patients, physicians should be aware that their advice might influence patients’ decisions away from their preferred treatment option. They should encourage their patients to identify their own preferences and help to find the treatment option most consistent with them.

Introduction

In recent years, there has been an increasing demand for a more active participation of patients in medical treatment decisions. ¹ , ² , ³ , ⁴ A more active participation of patients in decision making is considered appropriate, especially regarding ‘preference‐sensitive decisions’. ⁵ In this case, more than one medically reasonable treatment option exists yet the ‘optimal’ choice cannot be defined by objective criteria, but rather only by the individual patient’s values and preferences. ⁶ , ⁷ In such decisions, the patient has been suggested to play a major role in choosing the most appropriate treatment. ⁸

If doctors in these circumstances give recommendations based on their own preferences, they harbour the risk of deterring patients away from their preferred treatment option. This would not only counteract shared decision making but might also lead to poor satisfaction and low treatment adherence.

It was the aim of this study to investigate whether the advice given by physicians in pure preference sensitive decisions actually does pull patients away from their preferences and pushes them towards a decision that is against their values and preferences.

As studies from non‐medical research have shown that people more often accept a recommendation if the advice giver has greater confidence in his/her recommendation, ⁹ , ¹⁰ , ¹¹ we also studied the influence of physicians confidence on patients’ advice‐taking.

Methods

Study design

We used an experimental design incorporating a hypothetical decision scenario in which a treatment decision depended on the patient’s preferences. Our preference‐sensitive decision was a choice between two drugs that only differed in their side‐effects, not in their efficacy.

The experiment comprised two experimental conditions in which the level of physicians’ confidence (high vs. low) in their recommendation was varied. The assignment of patients to both conditions was random. The experiment was conducted by two trained experimenters.

Participants

Inpatients with multiple sclerosis and inpatients with schizophrenia were chosen as study participants, as preference‐sensitive decisions are often made in these cases (e.g. long‐term drug therapy with immunomodulatory/antipsychotic agents). Long‐term treatment is indicated in both groups and it is therefore especially important that patients’ preferences are taken into consideration to encourage treatment adherence. We chose to study two different diagnostic groups to test whether patients with different diagnoses differ in their advice‐taking behaviour.

Two investigators were in regular contact with the psychiatrists/neurologists from several wards of two neurological hospitals and two psychiatric hospitals in and around Munich, Germany, to screen patients suitable for the present experiment. Eligible for the study were inpatients between 18 and 65 years with the diagnosis of either multiple sclerosis (ICD‐10, G35) or schizophrenia (ICD‐10, F20/F23); patients had to speak German fluently and had to be judged by their physicians to be capable of consenting to the research. Patients were recruited shortly before discharge from the hospital. All patients gave written informed consent. The study was approved by the institutional review board (‘Ethikkommission der Technischen Universität München’).

Procedure

Patients were presented with a questionnaire describing a hypothetical decision scenario (see Fig. 1). Patients were told that the decision to be made was only a hypothetical one and asked to imagine that there were two different drugs available for the treatment of their disease (multiple sclerosis or schizophrenia). These drugs characterized as differing only with respect to their side‐effects (abdominal pain vs. headache) but not with respect to their efficacy. Patients were asked to indicate which of the drugs they would prefer. After having indicated their preference, patients were told by the experimenters that their physician had been asked prior to the experiment which of these two drugs he or she would recommend to the patient. The experimenters then supplied the patients with a written form, stating their physician’s recommendation. In 50% of the cases, it was stated that the physician was highly confident about his recommendation while the other half received a statement in which the physician expressed little confidence in his recommendation.

Figure 1.

 Procedure.

In fact, the recommendations were from the experimenters and not the patients’ physicians. We chose this procedure to get reactions from the patients likely to be closer to patient behaviour in the consultation.

As we wanted to investigate whether recommendations can pull patients away from their preference, the recommendation from the ‘physicians’ was always against the patients’ choice. Thus, when patients selected ‘drug A’ then ‘drug B’ was recommended and vice versa. After having received their physician’s recommendation, patients were requested to make a final decision (i.e. to make a choice between the two drugs).

Patients were asked to answer additional questions to identify factors that might have influenced their advice‐taking behaviour. After indicating their preferred drug but before receiving their physician’s recommendation, they were asked the question ‘How confident are you about your preference?’ and afterward they were asked, ‘How much do you trust your physician regarding this decision?’ After making their final decision, patients were asked ‘How satisfied are you with your decision?’ In all questions 10‐point Likert scales from 0 = minimum to 10 = maximum were used. To test whether patients paid attention to experimental manipulation (‘manipulation check’), they were asked ‘How confident was the physician about his or her recommendation?’ (response options: ‘confident’ or ‘not confident’) and ‘Did your physician recommend the same drug as your initial preferred choice?’ (response options: ‘physician recommended the other drug’ or ‘the same drug’).

Furthermore, patients supplied their age, gender, duration of the illness and level of education. To avoid any harm from our experiment to the patients (e.g. compromising trust in physicians) experimenters were instructed (and also provided with a detailed manual) to finish the experiment with an extensive debriefing in which it was cleared up that the patient had participated in a mind game or imaginational scenario. It was stated to every patient that the decision made in the experiment had no influence on their further therapy. Experimenters were advised to extend the debriefing until they were sure the patient had understood the hypothetical nature of the decision‐making scenario. According to the experimenters, no complaints about the experimental procedure were raised by study participants.

Note: In our manuscript, we do not differentiate between the terms ‘advice’ and ‘recommendation’ and use, as most studies do, both terms synonymously. ¹²

Statistical analysis

The main outcome measure was whether patients followed their physician’s advice or not (= patient’s advice‐taking behaviour). Thereby patients’ preferences based on drug information alone, served as a baseline.

Univariate comparisons of both diagnostic groups and also of the two experimental groups were assessed by chi‐square analysis and unpaired t tests for the following variables: sociodemographic and clinical variables (patient age, gender, education, duration of illness and length of hospital stay), patient advice‐taking behaviour, manipulation checks regarding experimental conditions and regarding the physician’s contrary recommendation, patient confidence about their initially preferred treatment option, patient trust in their physician and patient satisfaction with the final treatment decision. Additionally, all these variables were described as frequencies and percentages or as means and standard deviations.

To examine factors influencing the acceptance of physician recommendations, a backward stepwise logistic regression analysis was conducted. Experimental condition (physician’s confidence high vs. low), patient confidence in the initially preferred treatment option and patient trust in the physician as well as patient demographics and clinical variables were entered (patient diagnosis, age, gender, education, duration of illness and length of hospitalization). Four cases were excluded because of missing values.

The level of statistical significance was set at P = 0.05 and all tests were two‐tailed. Data were analysed using spss^® Version 16.0 (SPSS Inc., Chicago, IL, USA).

Results

Demographics and clinical variables

There was no significant difference between participants with schizophrenia (N = 102) and participants with multiple sclerosis (N = 101) as well as between patients in the ‘low‐confidence condition’ and patients in the ‘high‐confidence condition’ regarding age, gender, duration of illness and education, except for length of stay in the hospital (see Table 1).

Table 1.

 Patient demographics and clinical variables

	Patients with schizophrenia (N = 102)	Patients with multiple sclerosis (N = 101)	Difference between patients with multiple sclerosis and patients with schizophrenia	Difference between experimental conditions (separated according to diagnosis)
Female sex, No. (%)	56 (55%)	67 (66%)	n.s.*	n.s.*
Mean (SD, median) age (years)	36.5 (11.42, 34.50)	38.4 (8.29, 39.00)	n.s.†	n.s.†
Education
None or elementary school	33 (32%)	38 (38%)	n.s.*	n.s.*
Secondary school	35 (34%)	33 (33%)
Grammar school (Higher secondary school)	34 (33%)	30 (30%)
Mean (SD, median) illness duration (years)	9.8 (9.18, 7.00)	9.9 (8.58, 8.00)	n.s.†	n.s.†
Mean (SD, median) length of current hospitalization (days)	65.5 (46.08, 56.00)	9.9 (7.22, 8.50)	t = −12.03, P < 0.001†	n.s.†

*Chi‐square test.

^† t test.

Patient initial preferences regarding ‘drug A’ and ‘drug B’

Before having received the physician’s recommendation, 67% (n = 68) of the patients with schizophrenia and 59% (n = 60) of the patients with multiple sclerosis preferred ‘drug A’ which could cause abdominal pain (33%, n = 34 vs. 41%, n = 41 preferred ‘drug B’ which could cause headaches). The differences between both patients groups were not significant.

Patient advice‐taking behaviour

In total, 48% (n = 49) of the patients with schizophrenia and 26% (n = 26) of the patients with multiple sclerosis followed the advice of their doctor and thus chose the treatment option that went against their initial preferences (group difference: χ² = 10.83, P = 0.001).

Impact of physician’s confidence in their recommendation on patient advice‐taking behaviour

Patients with multiple sclerosis more often followed their physician’s advice in the ‘high‐confidence condition’ than in the ‘low‐confidence condition’; however, this difference failed to reach statistical significance (33% vs. 19%, χ² = 2.38, P = 0.12). In patients with schizophrenia, there was also no significant difference between the high‐ and the low‐confidence condition concerning advice‐taking (50% vs. 46%, χ² = 0.16, P = 0.69).

Manipulation check

Most patients paid attention to the manipulation: 81% (n = 83) of the patients with schizophrenia and 93% (n = 94; missing = 1) of patients with multiple sclerosis passed the manipulation check regarding ‘physician’s confidence in recommendation.’ Furthermore, the vast majority of patients were aware that their physician’s recommendation was against their preference (schizophrenia: 95%, n = 97); multiple sclerosis: 96%, n = 97, missing = 1).

Patient confidence about the initially preferred treatment option and patient trust in physicians

Patients with schizophrenia (mean = 7.1; SD = 2.43) were significantly less confident about their initially preferred treatment option than patients with multiple sclerosis (mean = 7.7, SD = 2.03; t = 2.05, P = 0.04).

Patient groups did not differ in how much they trust their physician in the actual decision between ‘drug A’ vs. ‘drug B’ (schizophrenia: mean = 6.3, SD = 2.63; multiple sclerosis: mean = 6.2, SD = 2.71; t = −0.28, P = 0.78).

Factors influencing patient advice‐taking behaviour

Three factors met the criteria for inclusion (P < 0.10) in the final logistic regression (see Table 2).

Table 2.

 Regression model for patient advice‐taking behaviour

Variable	Odds ratio	95% CI		P‐value
		Lower bound	Upper bound
Diagnosis
Multiple Sclerosis	Reference
Schizophrenia	3.03	1.51	6.08	0.002
Trust in physician	1.54	1.32	1.80	<0.001
Patient confidence about the initially preferred option	0.79	0.68	0.92	0.003

Patient diagnosis (P = 0.002), patient trust in the physician (P < 0.001) and patient confidence in their pre‐advice treatment choice (P = 0.003) predicted advice‐taking, with patients suffering from schizophrenia, patients with greater trust in the physician, and patients with less confidence in their initial choice being more likely to follow their physician’s advice.

Patient satisfaction with the final treatment decision

The two patient groups did not differ regarding their satisfaction with the final treatment choice (schizophrenia: mean = 6.4; SD = 2.71; multiple sclerosis: mean = 6.8; SD = 2.63). However, patients who accepted the advice of the physician were less satisfied with their decision (schizophrenia: mean = 5.7, SD = 2.59; multiple sclerosis: mean = 5.7, SD = 2.38) than patients who did not follow the physician’s advice.

Discussion

Almost half of the patients with schizophrenia and more than a quarter of the patients with multiple sclerosis followed their physician’s recommendation and were thereby steered away from their initial treatment preference. The recommending physician’s confidence in that choice (high or low) had no significant influence on patients’ willingness to follow the advice. However, patients’ confidence about their initial preference as well as patients’ trust in their physician were significant predictors. Patients who accepted their physician’s advice, and therefore changed their drug choice, were less content with the final decision than those who remained by their initially preferred option.

Our results show that doctors’ recommendations are not only ranked as important by patients ¹³ but may in many cases overrule patients’ preferences even in pure preference‐sensitive decisions. These findings imply that many patients are not willing to contradict their physician’s recommendation.

Looking at the factors influencing the advice‐taking behaviour of the patients in our study might allow for some explanation of this finding:

Participants with greater trust in their physicians more often followed the recommendations than those with less trust, a result that has also been found in non‐medical studies. ¹⁴ As trust is described as ‘the expectation that the person is both competent and reliable and will keep your best interests in mind’, ¹⁴ many patients might have just followed the (paternalistic) heuristic that a ‘doctor knows best’. ¹⁵

Patients who were less confident with their initial choice were more likely to follow their physician’s advice. Here many of those who expressed low confidence might have suffered from decisional conflict which has been shown to be common in patients facing health‐care decisions. ¹⁶

Advice‐taking behaviour was also strongly influenced by the patient’s diagnosis. Patients with schizophrenia were twice as susceptive to their doctor’s recommendations than patients with multiple sclerosis. As patients with multiple sclerosis have been shown to exhibit a greater desire to participate in treatment decision making than patients with schizophrenia, ¹⁷ this might explain why they less frequently followed the physician’s advice. Furthermore, patients with schizophrenia showed less confidence in their initial preference than those with multiple sclerosis. As mentioned above, lower confidence about initial preferences, increases the probability for advice‐taking. Finally, decisional capacity is often impaired in schizophrenia, ¹⁸ , ¹⁹ probably leading patients and physicians to be more accustomed to a paternalistic decision‐making style than in other diseases.

Although several psychological studies found a strong influence of the advisor’s confidence on advice‐taking, ¹⁰ , ¹⁴ this was not the case in our study. This difference might be attributed to the fact that in previous (psychological) studies, decisions with only one viable choice were used, whereas we used a preference‐sensitive decision. It is assumed that advice seekers use the confidence of the advice giver to infer the correctness of the advice. ¹¹ , ¹² As no objectively correct or incorrect decision existed in our study, physician confidence might not have had a major impact.

Of course, there may have been additional factors that could have influenced the advice‐taking behaviour in our study. It is possible that patients took the advice to share responsibility with their physician for the decision. ²⁰ , ²¹ Additionally, patients might have felt uncomfortable contradicting their physician and did not want to jeopardize the relationship. Furthermore, the belief that the physician had additional significant information that was not disclosed might have also influenced them to follow the physician’s advice. ²²

In summation, there are many factors influencing patients to accept their physicians’ advice, all of them quite understandable yet in some cases challenging the ideal of patients making informed (shared) preference‐sensitive decisions.

Thus, many patients maintain old paternalistic patterns even if the decisional situation (preference sensitive) encourages a high level of patient participation. Physicians should be aware of this finding and be particularly careful when giving suggestions.

That does not mean that physicians must not give advice on what they believe to be best for the patient.

Initially, physicians could encourage their patients to take part in the decision‐making process and to develop and articulate their own values and preferences regarding their medical care. ²³ Then, they could help their patients to identify the treatment options most suited to the patient’s preferences.

In a later step of the decision process, physicians may provide their treatment advice if patients ask for advice, are not able to make their own decision or do not want to participate in decision making. ²⁴ As there is evidence that people take advice for reasons of social pressure, ¹⁰ physicians should reassure patients that they do not expect their patients to follow this advice in any case. ²⁵ , ²⁶ If patients follow their physician’s advice without being truly convinced, there is a risk that they will not follow through on the advice, resulting in non‐compliance.

Limitations

Our study has several limitations. First, patients might behave differently in real‐life situations when interacting with their physician ‘in vivo’ instead of answering a hypothetical scenario, as in our experiment. We have tried to overcome this limitation by designing the vignette as realistic as possible and by studying patients instead of healthy volunteers. However, literature on this issue is inconsistent with regard to the criterion validity of hypothetical vignettes (e.g. ²⁷ , ²⁸ , ²⁹ ). Van Swol and Sniezek, ¹¹ for example, found that a direct interaction has more impact on deciders’ advice‐taking behaviour than a non‐personal, indirect interaction. Therefore, the effects found in our study might not be directly transferred to clinical practice.

Second, at the time of assessment, the patients suffering from schizophrenia had already been inpatients at the hospital for a longer duration than patients with multiple sclerosis, possibly resulting in stronger relationships with their treating physician. However, no correlation was found between length of hospitalization and trust in physicians and additionally, no influence was found in regression analysis for length of hospitalization and advice‐taking. Third, patients had no choice about receiving the physicians’ recommendation. There is evidence that solicited advice is more often followed than unsolicited advice. ³⁰ Nevertheless, as in real physician‐patient interactions, physicians often give recommendations without being asked explicitly by their patients.

Conclusion

Our study has shown a strong influence of physician advice on patients’ decision making, even in preference‐sensitive decisions where the best option by definition ought to be defined by patient’s preferences and not those of the physician. Patients who followed their physician’s advice were less satisfied with their decision than patients who did not.

Disclosure

The authors report no conflicts of interest relevant to the subject of the present manuscript. No source of funding or grant of any kind was used to finance this study.