Prediction of intended career choice in family medicine using artificial neural networks

Abstract

Background: Due to the importance of family medicine and a relative shortage of doctors in this discipline, it is important to know how the decision to choose a career in this field is made.

Objective: Since this decision is closely linked to students’ attitudes towards family medicine, we were interested in identifying those attitudes that predict intended career choice in family medicine.

Methods: A cross-sectional study was performed among 316 final-year medical students of the Ljubljana Medical Faculty in Slovenia. The students filled out a 164-item questionnaire, developed based on the European definition of family medicine and the EURACT Educational Agenda, using a seven-point Likert scale containing attitudes towards family medicine. The students also recorded their interest in family medicine on a five-point Likert scale. Attitudes were selected using a feature selection procedure with artificial neural networks that best differentiated between students who are likely and students who are unlikely to become family physicians.

Results: Thirty-one out of 164 attitudes predict a career in family medicine, with a classification accuracy of at least 85%. Predictors of intended career choice in family medicine are related to three categories: understanding of the discipline, working in a coherent health care system and person-centredness. The most important predictor is an appreciation of a long-term doctor–patient relationship.

Conclusion: Students whose intended career choice is family medicine differ from other students in having more positive attitudes towards family physicians’ competences and towards characteristics of family medicine and primary care.

Keywords::

• Medical students
• family medicine
• attitudes
• intended career choice
• artificial neural networks

KEY MESSAGE:

• Students who are likely to choose family medicine as their future profession have positive attitudes towards family physicians’ competences and characteristics of family medicine and primary care.

• The most important predictor of intended career choice in family medicine is a positive attitude towards a long-term relationship with the patient.

INTRODUCTION

Family medicine is a core discipline of primary medical care and the cornerstone of many health care systems in Europe. The European Union and non-EU countries have stressed the importance of family medicine and primary care in their official documents (1).

Population growth and an population aging will be the greatest drivers of expected increase in primary care utilization. Consequently, a larger number of primary care physicians will be necessary for the next decades in Europe (2–5). The lack of interest for family medicine among medical students and graduates of medical schools is an important issue in countries throughout Europe (6–10).

Reasons for choosing family medicine as a career choice are diverse. Various types of factors influence career choice in favour of family medicine (11–17): sociodemographic factors (female gender, growing up in a rural area, not having parents with a university degree) (12,15), personal characteristics (positive social self-image, personal ambition, wish to harmonize work and personal life) (11), and factors related to appreciation of specifics of family medicine in patient care (continuity of care, diversity, community orientation) (12,14,15). This last group of factors includes attitudes towards characteristics of family medicine. These factors may be essential in choosing a career in this field.

Attitudes towards the discipline are relevant, because they may be addressed through education. An undergraduate family medicine curriculum enables students to understand the value of family medicine in the health care system, and may influence their career choice (18–20). After the clerkship in family medicine, students report an improvement of their knowledge of and attitudes towards the discipline, but this was not found to be important in deciding for their future professional career (6,21).

Although various questionnaires have been developed to identify the career preference of medical students (11,12), most of them have dealt merely with the first two groups of factors. The European definition of family medicine/general practice, however, has never been used as a starting point; despite this definition being the basis of the EURACT Educational agenda, the reference standard for family medicine education in Europe (22). A questionnaire was devised that would assess attitudes of medical students towards characteristics of family medicine and competences of family physicians, defined in the European definition of family medicine/general practice.

The first aim of this study was to find out which attitudes towards the family medicine characteristics distinguish students who are likely to choose family medicine as their future profession from the ones who are not. The second aim of the study was to test the applicability of artificial neural networks in finding a subset of the most significant attitudes, regarding the career choice in family medicine.

METHODS

Design of the study

The study was designed as a cross-sectional survey including all medical students attending the family medicine clerkship in two study years (2010/11 and 2011/12) at the Medical Faculty in Ljubljana, Slovenia. Students completed a questionnaire on 164 attitudes on family medicine and stated their intention for a career choice in family medicine. In this context, attitude can be defined as a measure of preference towards a statement (regarding family medicine). The goal was to find attitudes with the highest predictive value for intended career choice in family medicine.

Study population

In our curriculum, the family medicine clerkship is an obligatory subject for sixth-year students. The programme lasts for seven weeks; students spend four days per week in the practice with their advisor, while one day per week is dedicated to teaching at the department. Details of the programme have been published elsewhere (6).

The analysis was performed on all 376 students attending the family medicine clerkship in two consecutive study years. Questionnaires were distributed by the teachers at the end of the clerkship.

Development of the questionnaire on attitudes towards family medicine

The European definition of family medicine was distributed among a group of 30 teachers and advisors of family medicine. They were asked to use their imagination in developing statements that would reflect attitudes towards the importance of each of the core competences listed in the EURACT Educational Agenda (22) and to provide a written response. The result was a list of 164 attitudinal statements about various features of family medicine. The statements were paraphrased by the authors and developed into a 164-item questionnaire.

Variables

The independent variables (‘items’) were 164 attitudes regarding family medicine. They were scored on a seven-point Likert scale (between 1 ‘strongly disagree’ and 7 ‘strongly agree’). To decrease the confounding effect of the ‘acquiescence response style’ (the tendency to constantly agree or disagree), approximately half of the statements in the questionnaire had a positive and half had a negative connotation.

The dependent variable was ‘intended career choice,’ measured with a five-point Likert scale (between 1 ‘very unlikely to become a family physician’ and 5 ‘very likely to become a family physician’). Answers were divided into three categories, representing the student's preference for family medicine:

1. unlikely to become a family physician (1 or 2 on the Likert scale);

2. neutral (3 on the Likert scale); and

3. likely to become a family physician (4 or 5 on the Likert scale).

The group of neutral students was excluded, and attitudes of the remaining two groups were compared. They may be also called ‘family medicine disinclined’ and ‘family medicine inclined.’

Data analysis

Basic data analysis

Descriptive statistics (mean values of Likert scale values ± SD) was used for the description of the independent variables (attitudes) and the dependent variable (intended career choice). The t-test was used to test the statistical significance of differences between both groups for individual items, and p ≤ 0.05 was considered to be significant. The χ² test was used to identify the influence of gender on intended career choice in family medicine.

Classification of students regarding intended career choice in family medicine

Pattern classification (or simply ‘classification’) treats the problem of assigning a class label to an observation based on the training set of data containing observations with known class memberships. In this way, it is possible to predict the class membership for a new observation, based only on its item values. A classifier is a rule that maps observation data to a class. So, it predicts to which class an object belongs. In this way it discriminates between two (or more) classes.

Artificial neural networks

Artificial neural networks (ANNs) are classification models, inspired by biological nervous systems (23,24). In comparison with logistic regression, ANNs are able to detect more complex (non-linear) relationships between independent and dependent variables. ANNs have been shown to improve the predictive value of classical statistics in many areas of medicine (25,26), but they have not been used yet for prediction of career choice in family medicine.

ANNs consist of many highly interconnected units or neurons, which together perform a complex input-output relationship. The complexity of this relationship is dependent on the size of the network (number of internal or ‘hidden’ neurons, H). ANNs are able to adapt their free parameters in a process called training, to realize desired input-output mappings; in our case between the Likert values of the items and the intended career choice. They can also predict intended career choice for new students based on their attitudes. We applied the multilayer perceptron (MLP), the most widely used type of ANN, trained with the back- propagation algorithm (24).

Step 3. Selection of items that most accurately predict intended career choice

In machine learning and statistics, the process of finding a subset of relevant independent variables for use in model construction is called feature selection. One of the most widely used methods is the sequential forward selection (SFS) (27). The model in this case is the ANN classifier (MLP). To find such a subset of items that most accurately predict intended career choice on the basis of known attitudes, the SFS procedure with classification accuracy of the MLP was implemented as the evaluation function. The SFS is a stepwise procedure, which starts with an empty set of items. At each step, the item that maximizes the classification accuracy of the MLP is found and joined to the set of items. In this way, items are ranked by importance.

The SFS procedure was performed for MLPs of different size (number of hidden neurons H), which affects complexity. The best items for different values of H are combined to give the best 31 items overall.

Ethical approval

The national ethical committee approved the study (17-4-2007). The number of the approval was 125/04/07.

RESULTS

Description of the study sample

None of the 376 students explicitly refused to participate in the study, but due to missing data about the intended career choice in family medicine, 60 questionnaires (16.0%) were excluded; 208 female (65.7%) and 108 (34.3%) male students remained. Table 1 shows the distribution of the students’ answers according to the intended career choice in family medicine. Since 115 students answered neutrally about their intended career choice (3 on the Likert scale), their data were excluded from further analyses.

Table 1. Preference for family medicine in 316 final-year medical students.

Students’ intention to become a family physician (n = 316)	Frequency	Percentage	Class
1 (Family medicine is not my career choice at all)	24	7.6%	Family medicine
2	80	25.3%	disinclined
3	115	36.4%	(Neutral)
4	83	26.3%	Family
5 (Family medicine is my career choice for sure)	14	4.4%	medicine inclined

The final dataset of 201 students (125 female, 62.2%) was divided into two classes: ‘family medicine disinclined’ (n = 104) and ‘family medicine inclined’ (n = 97), respectively. There were no differences between male and female students regarding their wish to become a family physician (χ² = 0.433, P = 0.496).

Attitudes that predict intended career choice in family medicine

Figure 1 shows the classification accuracies arising from adding the best item at each step in the case of a MLP of the lowest complexity. At the first step, the best item was ‘Working always with the same people is boring’ (item 41), yielding 70.1% classification accuracy alone. At the second step, the highest rise in accuracy was the result of adding item ‘Physicians educate by being a role model’ (item 104), etc. Obviously, Figure 1 shows that the largest differences in accuracy arose when adding the first six items.

Figure 1. Classification accuracy arising from adding the best item at each step of the SFS procedure (MLP with one hidden neuron, H = 1).

Figure 2 shows the course of classification accuracy of the MLPs of different complexities (number of hidden neurons H) during the SFS procedure. The best items for different values of H are combined to give the best 31 items overall.

Figure 2. Classification accuracy in MLP models of different complexity (regarding the number of hidden neurons H) when adding the first 20 items.

These are ranked and the first 12 of them are presented in Table 2. The best item regarding the discrimination between students about their preference for family medicine is item ‘Working always with the same people is boring.’ The other items correspond to attitudes to patient-centredness of care, to public health care and coherent health care system, and to teaching by role modelling. Mean values (SD) and the differences between both classes of students are also presented in Table 2.

Table 2. Twelve most important attitudes for discrimination between students regarding their preference for family medicine (analysed by SFS with ANNs, starting from a list of 164 attitudes, based on the European definition of family medicine and the EURACT Educational Agenda). The mean values and the standard deviations (SD) of both groups in the Likert scale are given in the second and in the third column.

Attitude (item)	Mean (SD) of family medicine disinclined (n = 104)	Mean (SD) of family medicine inclined (n = 97)	p	Connotation
1 Working always with the same people is boring (item 41)	3.50 (1.73)	2.25 (1.49)	< 0.001	−
2 The public healthcare system does not provide quality care (item 124)	3.08 (1.78)	2.21 (1.33)	< 0.001	−
3 Physicians educate by being a role model (item 104)	5.64 (1.41)	6.16 (1.12)	0.004	+
4 Trivial health problems are often the ticket to deal with serious health problems (item 31)	4.52 (1.58)	5.09 (1.39)	0.007	+
5 Without good communication one cannot work well in the practice (item 53)	6.23 (1.11)	6.71 (0.68)	< 0.001	+
6 It is important to know insurance policies while working in the practice (item 121)	6.15 (1.13)	6.52 (0.77)	0.008	+
7 Home visits are an outdated form of physicians’ work (item 86)	2.71 (1.55)	1.91 (1.20)	< 0.001	−
8 Referrals are a method of achieving equitable access to the secondary level (item 78)	5.24 (1.47)	5.81 (1.23)	0.003	+
9 A physician must be familiar with the principles of the health care system (item 140)	5.80 (1.33)	6.38 (0.74)	< 0.001	+
10 Preventive work is the key to health (item 109)	5.70 (1.33)	6.16 (0.95)	0.005	+
11 The psychological aspect of the disease is often just as important as the physical aspect (item 98)	6.04 (1.09)	6.46 (0.87)	0.002	+
12 Physicians should lead the patients through the health care system (item 17)	4.67 (1.46)	5.27 (1.25)	0.002	+

DISCUSSION

Main findings

Using an innovative approach of classification with ANNs applied in the SFS procedure, 31 attitudes were found, which best predict an intended career choice in family medicine in final-year medical students. Positive attitudes towards ‘understanding of the discipline,’ ‘working in a coherent health care system’ and ‘person-centredness’ are the key attributes of the family medicine inclined students. Specifically, 12 most important attitudes were listed, some of which not having been identified in the literature before, such as ‘working in a coherent health care system’ and ‘positive attitudes toward public health care.’ This list creates a picture of a prototype student who is likely to choose family medicine as a career option. Such a student appreciates a long-term physician–patient relationship, which enables patient-centred care, taking into account prevention and treatment of physical and psychosocial problems. The student is aware of the ethical concept of equity and knows the rules of the health care system, which are both fundamental for socially oriented and high-quality health care.

Strengths and limitations

This study is the first study using the European definition of family medicine and the EURACT Educational agenda as starting points for the questionnaire on attitudes towards family medicine. The study used artificial neural networks in predicting career choice in family medicine for the first time. ANNs yield higher classification accuracies than logistic regression methods (26).

There are also some limitations. First, the study was conducted only among students of one out of two medical faculties in the country. It is possible that the population of students at the second (younger) medical faculty are different in characteristics and career plans from the population of students at the faculty with a long-lasting tradition (28). Second, given the nature and structure of the health care system, culture and the role of family physicians in Slovenia, the findings may not be directly applicable to other countries. In Slovenia, there is only one national health insurance company, all citizens have at least basic health insurance, and family physicians, who are gatekeepers, treat a fixed list of patients—they usually have a long-term relationship with their family physician (29). A third limitation was that almost one fifth of students did not answer the question about intended career choice in family medicine. These students may be less motivated for family medicine, and it is possible that the percentage of family medicine inclined students in this group is smaller than in the analysed sample.

Interpretation of the study results in relation to existing literature

The proportion of students who were certain to choose family medicine as a career choice (4.4% or 14/316; answer 5 on the Likert scale) was smaller than in Germany (7%) (11) and Spain (11.4%) (16).

Six items were especially important in predicting intended career choice in family medicine (Table 2, items 1–6). The most important factor is the opinion that working with the same people (patients) is not boring, which we interpreted as a positive attitude towards long-term relationships with patients. Long-term relationships are typical for primary care, and their value was an important predictor of career choice in some other studies as well (14,15). However, long-term doctor–patient relationship was an inhibiting factor for career in family medicine due to the responsibilities related to long-term care (30).

Family medicine inclined students are more prone to believe that work in family medicine can be of high quality—not only in hospital settings, but also in primary care—understanding the quality of care at the primary level as a broader concept and taking into account individual level and level of society (31,32). This is the second most important factor.

The third factor is a positive attitude towards a family physician as a role model. In our curriculum (and many others), teachers of family medicine are often seen as positive role models for students (6). In another study, general practitioners’ professional self-perception was more positive than the way specialists working in hospitals view general practitioners (17). We see here a potential that education can offer: negative comments about family medicine in the teaching of some clinical specialists influence medical students’ decisions on a career in family medicine; however, having family medicine role models from early on in medical school might have the opposite effect (33).

Further important attitudes are related to societal and social orientation of the physician's role. Students who prefer family medicine are more societal oriented and also have more interest for social problems than students intended to become clinical specialists (12).

Implications for medical education

When identifying new factors that predict preference for family medicine, it is important to see which of them can be addressed in the curriculum (34). It seems that the attitude towards a long-term doctor–patient relationship is difficult to address in the curriculum.

There are different results regarding different curricula's ability to influence career choice in family medicine (6,21), but there is an agreement on the importance of role models during medical school (14) and on their role in shaping students’ attitudes.

Implications for future research

It would be interesting to observe how young students’ attitudes towards family medicine evolve throughout their studies, how accurately the intended career choice at the end of the family medicine clerkship predicts the actual career choice, and how accurately our reduced list of attitudes predicts the actual career choice.

CONCLUSION

This study identified medical students’ attitudes towards family medicine associated with intended career choice in family medicine in the final year of medical school. Positive attitudes towards family physicians, a long-term doctor–patient relationship and the characteristics of primary care and the health care system are important predictors of intended career choice in family medicine. Also the artificial neural networks can be successfully used in the prediction of an intended career choice in family medicine.

ACKNOWLEDGEMENTS

The authors should like to thank the team of the Ljubljana Family Medicine Department for their valuable input in the creation of ideas for the questionnaire; the students who fulfilled the questionnaire; and Ana Artnak, secretary of the department for her administrative support.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.