What patients consider to be a 'good' doctor, and what doctors consider to be a 'good' patient.

BACKGROUND
From a patient's point of view, an 'ideal' doctor could be defined as one having personal qualities for interpersonal relationships, technical skills and good intentions. However, doctors' opinions about what it means to be a 'good' patient have not been systematically investigated.


AIM
To explore how patients define the characteristics of a 'good' and a 'bad' doctor, and how doctors define a 'good' and a 'bad' patient.


MATERIAL AND METHODS
We surveyed a cohort of 107 consecutive patients attending a community teaching hospital in February 2019, who were asked to define the desirable characteristics of a good/bad doctor. Additionally, a cohort of 115 physicians working at the same hospital was asked to define the desirable characteristics of a good/bad patient. Responses were subjected to content analysis. Simultaneously, an algorithm in Python was used to automatically categorize responses throughout text-mining.


RESULTS
The predominant patients' perspective alluded to desirable personal qualities more importantly than proficiency in knowledge and technical skills. Doctors would be satisfied if patients manifested positive personality characteristics, were prone to avoid decisional and personal conflicts, had a high adherence to treatment, and trusted the doctor. The text-mining algorithm was accurate to classify individuals' opinions.


CONCLUSIONS
Ideally, fusing the skills of the scientist to the reflective capabilities of the medical humanist will fulfill the archetype of what patients consider to be a 'good' doctor. Doctors' preferences reveal a "paternalistic" style, and his/her opinions should be managed carefully to avoid stigmatizing certain patients' behaviors.

A n 'ideal' doctor could be defined as one having personal qualities for interpersonal relationships, proficiency in knowledge, technical skills and good intentions 1 . But beyond these considerations, most doctors are good doctors in the eyes of most patients, and complaints are usually related only to the fact that doctors and patients do not always agree on priorities 2 . Persons who have recently experienced medical care generally give less critical ratings to doctors 3 , and while older patients are tolerant with medical 'paternalism', younger patients tend to highlight doctors' skills in communication and in sharing decision-making 4 . Patients' perceptions regarding doctors' qualities have been explored with various structured questionnaires [5][6] and occasionally through a qualitative approach with open-ended questions 7 . Though the results of structured questionnaires are easier to analyze, the open enquiry format allows to gather information on patients' beliefs and expectations, not ruled by pre-established categories.
In contrast, doctors' opinions about what it means to be a 'good' patient have not been systematically explored. From the physician's point of view, the implicit characteristics of a 'good' patient would include: doctor-trusting, docility to avoid decisional conflicts, obedience to doctor's commands, high adherence to treatment and continuity of care 8 . Since patients and doctors may have different views or priorities to characterize desirable individual's features, we designed this study with the aim of exploring how hospitalized patients define the characteristics of a 'good' and a 'bad' doctor, and how doctors define a 'good' and a 'bad' patient.

Materials and Methods
We surveyed a cohort of consecutive patients attending a community teaching hospital in February 2019. The group corresponded to patients hospitalized in a general internal medicine or surgical ward, or in a day-care hospital unit. All patients hospitalized during the study period were considered. Exclusion criteria were dementia, aphasia, no consent or unwillingness to participate, or any difficulty with the Spanish language that prevented them from understanding or answering a questionnaire. Patients were also stratified to evaluate the effect of age, gender, level of education, and perceived health status on responses. Patients invited to participate were interviewed by a researcher in order to inform them that this study would explore their definitions of what made a good/bad doctor, that they were not asked to evaluate the care they were receiving during their present hospital stay, and that their responses would not impact on their care. The protocol was evaluated and approved by the Institutional Review Board.
A qualitative approach with an open-ended questionnaire was used to allow the interviewee to give his/her own free hardly influenced answers. Questions were formulated as follows: "According to you, what is an ideal doctor, a doctor you would like to be treated by? How would describe him/ her?", and otherwise "According to you, what is a bad doctor, a physician you would not like to be treated by? How would you describe him/her?" Complete confidentiality was guaranteed and responses were rendered anonymous. Patients' responses were subjected to content analysis performed by a psychologist and a physician (CM and  (Table 1). This list was subsequently refined and enlarged to include unforeseen answers. Categorization of ambiguous responses was discussed and disagreements were solved by consensus. Additionally, a cohort of physicians working at the same hospital was asked to anonymously respond a similar open-ended questionnaire in order to define the desirable characteristics of a good/bad patient. The questions' structure was: "According to you, what is an ideal patient, a patient you would like to treat? How would describe him/her?", and otherwise "According to you, what is a bad patient, an individual you would not like to treat? How would you describe him/her?" Due to lack of previously described categories to classify doctors' responses, we defined by consensus the list of possible characteristics that a 'good' patient would present (Table 2). In this case, the categories of a 'bad' patient were initially defined as the reverse image of the good one. Similarly to the patients' survey analysis, relevant topics obtained from responses were grouped into categories encompassing the characteristics of good/bad patients as communicated by doctors. Physicians' demographic characteristics such as age, gender, medical specialty, and years from graduation were also recorded.
Simultaneously, an algorithm written in Python 3.7.0 (Python Software Foundation. 2001-2018) was used to automatically categorize patients' and physicians' responses by searching selected keywords throughout the free texts (text-mining) (Appendix). Finally, human (psychologist/physician) versus algorithm-based analysis was compared.
Qualitative variables were expressed as absolute values and percentages, and continuous variables as mean, range and standard deviation (SD). Comparison between categories was made with c², using the 2-tailed Fisher's exact test when cell expected values were ≤ 5. IBM SPSS 23.0 Statistics (IBM Corporation, Armonk, NY) was used for statistical analysis.

Patients' responses
One hundred and seven patients completed the survey. Mean age was 47.1 years (SD 16.2, range 18-77) and 57.0% were women. Regarding patients' educational level, 8.4% had attained primary school level, 41.1% secondary level, and 50.5% tertiary/university level. Considering patients' self-rated health status, 45.8% of respondents declared a very good or excellent health status, 45.8% referred to it as good, and 8.4% as fair. Table 3 summarizes the patients' responses for what they considered was a 'good' doctor, according to the human and algorithm-based analyses. Comparison between the two analytical methods showed no statistical difference in most categories, except for the "communicational skills" dimension that was more frequently identified by the algorithm than by the human analyzer. Literal descriptions of a 'good' doctor are referred in the following responses. Example 1: "... he/she is respectful, attentive and sincere ... he/she explains everything about the procedures I am going to receive". Example 2: "... he/she has vocation for service... efficient... responsible". Example 3: "... he/she generates empathy, confidence... he/she is in a good mood, which conveys that you are not another number".
Undesirable physicians' features as denoted by the same patients are included in Table 4. Comparison between the two analytical methods showed no statistical differences in most categories, except for being "unskilled in communication" that was more frequently indentified by the algorithm than by the human analyzer. Literal descriptions of a 'bad' doctor are referred in the following responses. Example 1: "... he/she seems indifferent and treats a patient simply as a thing..." Example 2: "... he/she is always in a hurry...does not understand that his/her patient is a person with a history and emotions, not just a body to treat...does not listen to his/her patient". Example 3: "... he/she has not the capacity to understand the bad moment the patient is experiencing...". Regarding gender differences, female patients tended to prefer "sensitive to emotions" physicians, when compared with male respondents (51% versus 35%, p = 0.098). Younger patients below the mean age (47.1 years) considered "communication skills" as a desirable doctor's trait when compared with older individuals (41% versus 19%, p = 0.013). Regarding Patient does not know 0 (0.0) 0 (0.0) 1.000 As more than one response was possible, the total is higher than 100%. Interested only in money (works for money; lack of dedication) 0 (0.0) 0 (0.0) 1.000 As more than one response was possible, the total is higher than 100%.
Good and bad doctors and patients -R. A. Borracci et al educational level, patients who attained tertiary or university level preferred "sensitive to emotions" physicians (56% versus 32%, p = 0.014), whereas those who attained primary or secondary educational level preferentially chose "positive personality traits" (60% versus 28%, p < 0.001). Finally, when patients were divided by self-rated health status, individuals reporting a very good or excellent health status tended to consider the "lack of scientific proficiency" an undesirable characteristic for defining a "bad" doctor (18% versus 5%, p = 0.061). The rest of possible comparisons did not show statistical differences or trends among patients' demographic characteristics (Figure 1) (see also Tables 7 to 11bis in the supplementary material).

Physicians' responses
One hundred and fifteen physicians completed the survey. Mean age was 40.4 years (SD 11.5, range 26-65) and 53.0% were women. Regarding medical specialties, 40.9% performed a surgical specialty and the rest a clinical one. Mean time from graduation was 14.5 years (SD 9.7). Table  5 summarizes the physicians' responses for what they considered was a 'good' patient, according to the human and algorithm-based analyses. Comparison between the two analytical methods showed no statistical differences in all categories. In order to simplify the analysis, obedience to doctor's commands and adherence to treatment categories were grouped into only one category. A 'good' patient was defined by individual physicians with the following responses. Undesirable patients' features as referred by the same physicians are included in Table 6. Comparison between the two analytical methods showed no statistical differences in most categories, except for the "conflicting attitude" and "sharing decision-making" dimensions that were more frequently indentified by the algorithm than by the human analyzer. Literal descriptions of a As more than one response was possible, the total is higher than 100%. Sharing decision-making (uncommitted with self-care) 0 (0.0) 24 (20.9) < 0.001 As more than one response was possible, the total is higher than 100%. Physicians' opinions were divided by age below and over the mean value of 40.4 years. Older doctors preferred patients "obedient to doctor's commands and suggestions" (63% versus 35%, p = 0.003), whereas younger physicians favored patients with "positive personality traits" (78% versus 54%, p = 0.008) when choosing among categories for "good" patients. No age-based differences were observed when selecting categories for "bad" patients.

Discussion
In this qualitative study, we explored the definitions of what it means to be a 'good' or a 'bad' doctor from a group of hospitalized patients' point of view. Patients' opinions on doctors' positive and negative characteristics were contrasted with the definitions of what it is to be a 'good' or a 'bad' patient from the doctors' perspective.
At a first glance, features selected by patients to define a 'good' or a 'bad' doctor showed that the 'bad' doctor cannot be defined as the reverse image of the 'good' one, since 'bad' doctors were defined more by their negative personality characteristics than by sensitivity to emotions. Although positive/negative personality traits and sensitivity/ insensitivity to emotions were in the foreground in both 'good' and 'bad' doctors, the psychologist and the physician who analyzed the responses recognized that these characteristics were difficult to differentiate from each other.
One approach to defining a 'good' doctor equates the skills of a dedicated scientist and a medical humanist 2 . In the current study, scientific proficiency was ranked in the third place after more humanistic characteristics as sensitivity to emotions and positive personality traits to define a 'good' doctor.
The communication skills in a 'good' doctor were claimed by almost a third of the patients, although the text-mining analysis found them in almost half of respondents. A review of the literature on primary care patients' perspective about how patients want their doctor to communicate revealed specific expectations to be met in medical encounters 9 .
In a recent survey, patients were asked to respond in two words their feelings about the doctor's interview immediately following the clinic visit 10 . Positive words more frequently chosen were: knowledgeable, caring, professional, excellent, and competent. Negative words frequently chosen were: rushed, busy, hurried, uncaring, rude, unconcerned, arrogant, uninterested, and condescending. In this case, positive words predominantly referred to scientific proficiency, whereas negative adjectives suggested poor personality traits. In our study, the positive/negative personality characteristics and the emotional sensitivity/ insensitivity categories ranked both in the first places to define a good/bad doctor.
Regarding what physicians considered to be a good/bad patient, most common categories selected were the positive/negative personality traits and the adherence to doctor's suggestions. At least for the two top categories, 'bad' patients could be defined as the reverse image of a 'good' one. A conflicting attitude of the patient or his/ her family was indicated as an unfavorable characteristic by almost one-fifth of physicians, but this frequency increased to one-third after using the algorithm-based method.
Nearly one-fifth of physicians considered patient trust a desirable characteristic, and less than 10% cared about the concept of sharing decision-making, although the text-mining analysis duplicated this last proportion, at least for those considered 'bad' patients. Wrede-Sach et al. 11 found that when confronting "self-determined" to "doctor-trusting" patients, it was not easy for doctors to anticipate the desired level of patient involvement in their healthcare shared decision-making.
A longitudinal study of student experiences in clinical learning conducted by Sointu 12 concluded that 'bad' patients were considered mainly to have wrong priorities, little knowledge, and were difficult to deal with, while 'good' patients were active, compliant and knowledgeable. Labeling patients as 'good' or 'bad' has been also recognized to affect their abilities to make appropriate decisions. 13 Free-form text analysis can be performed by almost instant automated information extraction using text-mining methods. However, because of the severe implications that errors may have in healthcare, potential benefits of human/natural language technologies must be carefully evaluated 14 .
In the current study we explored a simple algorithm to automatically categorize patients' and physicians' responses by searching selected keywords throughout the free-form texts. Later, algorithm outcomes were compared with the classifications done by the psychologist/physician researcher. Globally, categorization of patients' opinions done by the algorithm was accurate for most groups.
This study has some limitations. First, although we did not try to make a difference between a 'poor' and a 'bad' doctor, surveyed patients included these categories together. For instance, a 'poor' doctor is generally credited with good intentions but inadequate knowledge or skills required for the job, while a 'bad' doctor, however skilled, is one with bad intentions, undesirable values, or serious defects of moral agency 2 . Second, since all the respondents were not critically-ill patients, most of them reported a good or very good self-rated health status; hence, this situation should be considered to interpret the current findings. Third, these findings are difficult to generalize since the patients' and doctors' responses are heavily influenced by our social and cultural context. Finally, until further investigation, the free-form text-mining analysis should be considered an exploratory approach.
In conclusion, this survey-based qualitative study gathered local information on patients' and physicians' opinions about what they considered is a good/bad doctor or patient, respectively. Although it was initially hypothesized that patients and doctors would have different views or priorities to characterize desirable individual's features, personality characteristics were in the foreground in both patients and physicians' selections. In addition to the human-based analysis, a text-mining algorithm was accurate to classify individuals' opinions into the different preset categories. Ideally, fusing the skills of the applied scientist to the reflective capabilities of the medical humanist will fulfill the archetype of what patients consider to be a 'good' doctor. Furthermore, doctors would be satisfied if patients manifested positive personality characteristics, were prone to avoid decisional and personal conflicts, had a high adherence to treatment, and trusted the doctor. These preferences reveal a "paternalistic" physician style, and his/her opinions should be managed carefully to avoid stigmatizing certain patients' behaviors.