Health care spending worldwide continues to increase and now accounts for approximately 17% of the gross domestic product in the United States, 9.8% in the Organisation for Economic Co-operation and Development (OECD) countries, and 7.5% in Israel [1]. Most experts consider the level of health care spending in the United States unsustainable [2]. Health economists identify unnecessary diagnostic and screening tests as a primary driver of this spending [2-5]. Moreover, many studies have shown that overuse neither benefits health care nor health outcomes [6-9] and may have adverse effects [10], leading to more unnecessary tests and treatment [11]. Major attempts to prevent the overuse of health services (HSs) include the British Medical Journal series Too Much Medicine [12], established in 2002, and Choosing Wisely campaigns [13], established in 2012. Increasing interest in this area is also reflected by the growing number of books, literature, and articles in the mainstream media [14].

In this study we chose to investigate medical practice variation (MPV), coined by the Dartmouth research group [15,16]. While MPV is based on a relative comparison between providers (Figure 1), overuse and underuse definitions call for a comparison between the individual provider practice and standard of care (absolute comparison) based on a “gold standard” or guideline recommendations [16,17]. Both overuse and underuse (eg, patients not receiving optimal care and resources used inefficiently) have negative consequences [14] and can contribute to MPV [15]. MPV is associated with poorer health outcomes, increased costs, disparities in care, and burden on medical systems [14,17-21]. Adopting the policy aimed to reduce variation is a central theme of quality management that has begun with industrial production and was recently adopted in medicine practice [19,22,23].

Not all MPV has pernicious effects. One should distinguish between unwarranted and warranted MPV. Warranted MPV reflects patient-centered care as it takes into account patient differences such as clinical or patient-preference differences [20,24]. Unwarranted MPV is caused by many factors such as variable access to resources or differing physician opinions and preferences [15,16]. In our study, we try to focus on factors contributing to the unwarranted MPV.

MPV causes can be divided into 3 main domains: patient characteristics, health care system characteristics, and physician characteristics [25,26]. Existing research suggests that variation is mainly attributable to patient characteristics, rather than to physician, or clinic [27-31]. Patient-related factors frequently studied in primary care [32] included age [33,34], type and severity of illness [33-35], socioeconomic status [36-39], ethnicity/race [40-42], and expectations of treatment effect [34,35,43,44]. Clinic factors included size [38,39,45,46], workload [46-48], funding method [49], services availability [35,50,51], services cost [52,53], and rural/urban location [15,34,54-56].

Factors related to the primary care physicians can be divided into demographic/professional and psychological characteristics. Demographic/professional characteristics include age, gender [33,42,46,47,57,58], specialty [38,58], area of expertise [36,47,57,59], and years of practice [33,46,58,60,61]. The psychological characteristics are discussed further on in the manuscript.

It appears that the major part of the variation is unexplained [31,62,63], that is, there are more influential factors that were not adequately estimated and researched so far [64], such as system-level and physician-level psychological characteristics. Assessing the extent of the variance that can be explained by the physicians’ psychological characteristics is essential because it will allow us to develop behavior modification tools that might help in reducing MPV [25]. Targeted interventions that address these variables might successfully optimize test ordering. The physician psychological characteristics include personality [65], attitudes [66], or behavior [67].

There is no generally agreed upon definition of personality, yet it is defined as person’s stable feelings, thoughts, and behavioral patterns [68]. The Big Five dimensions of personality are openness, conscientiousness, extraversion, agreeableness, and neuroticism [69]. Attitude refers to one’s opinions, beliefs, and feelings about aspects of his/her work environment. Two job attitudes have the greatest potential to influence how people behave at work: job satisfaction (feelings people have toward their job) and organizational commitment (the emotional attachment people have toward the company they work for) [68].

Compared with personality and attitudes, behavior is less abstract and more measurable, objective, and quantified and also it encompasses the other variables as it is a derivative of them [67,70]. Therefore we chose to focus on the physicians’ behavior characteristics. Four types of behaviors have been extensively studied in the organizational behavior literature: job performance, organizational citizenship behaviors, absenteeism, and turnover [68]. Our research, which studies the referral rates of the physicians, is related to the job performance behavior. Job performance refers to the success that one accomplishes in the tasks listed in his/her job description. Factors related to a physician’s job performance and medical decision making are the way s/he is treated at work, cognitive shortcuts (heuristics), the level of stress experienced at work, work attitudes to risks, and emotion [68,70].

Personal behavioral characteristics studied to date with respect to unwarranted variation included risk attitudes [58,71-74], adherence to treatment guidelines [50,75-78], empathy [79], and fear of malpractice [80]. We believe that the comprehensive approach where we will investigate the effect of risk aversion, stress due to uncertainty, fear of malpractice, empathy, and burnout will bring a higher level of inference as each can be reliably measured and may uniquely and independently account for significant MPV.

While most MPV research has focused on secondary and tertiary care in health regions and hospitals [15], this study examines MPV across primary care physicians. Primary care accounts for 14% of all health care spending on average across OECD countries with patient–physician consultations accounting for the majority (55% in the United States; 90% in the UK) [96]. Determining what accounts for MPV in primary care can help to develop targeted approaches for preventing unnecessary tests and treatment, better care coordination, cost containment, and improved health outcomes.

In 2019, Israel was ranked the 10th healthiest country in the world by Bloomberg rankings, out of 169 countries [97]. The National Health Insurance Law of 1995 mandates all citizens resident in the country to join 1 of the 4 official not-for-profit health maintenance organizations, which are prohibited by law from denying any Israeli resident a membership [98]. The study is placed in the Southern District of Israel, the Negev, and includes physicians and patients of the Clalit Health Services health maintenance organizations, the largest health insuring organization in Israel (4.5 million insurees). Clalit Health Services is the largest health care provider in the area, covering approximately 70% of a population of 730,000 residents in the Negev.

In this study we aimed to analyze HSs in the primary care in situations where the physician has the freedom of action to decide whether to utilize them [99], that is, clinical scenarios with discretionary decisions [100]. For instance, referring a patient with ST-elevation myocardial infarction to the hospital is not a discretionary decision as the physician’s findings reflect an undoubtable urgent condition of myocardial infarction which can be cured in a catheterization room. But ordering a thyroid-stimulating hormone (TSH) test for a patient with generalized weakness is discretionary, because weakness is a nonspecific symptom which can be caused by many factors such as an infection, anemia, inflammation. For these discretionary HSs, different choices carry different benefits and risks and therefore we believe that physicians will differ in the decisions [101,102], based on their knowledge, experience, beliefs, and thoughts. We hypothesize that the derivative of these components, the physician personal behavioral patterns, can significantly influence the utilization of the services in this category. Therefore, we identified the 16 most frequently utilized HSs in primary care that can be discretionary.

This paper describes a 3-phase research protocol (Figure 2) of MPV of primary care physicians across 16 HSs in the largest health care network in Southern Israel (Clalit Health Services) between 2003 and 2017.

The study’s objectives are to (1) describe the variation of HSs referrals among primary care physicians; (2) derive the unexplained variance after the adjustment for patient, physician, and clinic characteristics; (3) assess the extent to which the personal behavioral characteristics of the primary care physicians contribute to the explanation of the unexplained variance; and (4) identify the potential targets for behavioral modification interventions to reduce the variation in practice patterns and thus improve health care, decrease costs, and prevent disparities in care.

We hypothesize that the physicians’ personal behavioral characteristics such as risk aversion, stress due to uncertainty, fear of malpractice level of empathy, and burnout are affecting the decision-making process as demonstrated by the different levels of their HSs referrals.

This research was done without patient involvement. Patients were not invited to comment on the study design and were not consulted to develop patient-relevant outcomes or interpret the results. Patients were not invited to contribute to the writing or editing of this document for readability or accuracy.

The first phase will delineate the variation in primary care practice over a wide range of services and long follow-up period, the second will examine the 3 determinants of variation (ie, patient, physician, and clinic characteristics) and attempt to derive the unexplained variance. In the third phase, we will assess a novel component that might contribute to the previously unexplained variance—the physicians’ personal behavioral characteristics such as risk aversion, fear of malpractice, stress from uncertainty, empathy, and burnout.

Data will be collected from the computerized medical records of Clalit Health Services and will include (1) patient data (age, sex, marital status, residence type [urban, rural, or nomadic], number of annual visits by community physicians, background diseases, and socioeconomic status according to address); (2) primary care physician data (age, sex, years of active practice, area of expertise, specialty, country of birth, familial status, ethnicity, country where medical studies were completed, and number of insured patients in the physician’s unit); (3) clinic data (number of insured patients per doctor, number of physicians per 1000 patients, number of annual visits, location [rural/urban]); (4) HS data (annual referral number per physician per clinic for bone scan, computed tomography, chest x-ray, magnetic resonance imaging, Holter ECG, echocardiogram, stress test, transesophageal echocardiography, vitamin B₁₂ test, vitamin D test, TSH test, hemoglobin test, carcinoembryonic antigen test, prostate-specific antigen test; specialist consultation visits for rheumatology, pulmonary, and neurology; and emergency department visits due to chest pain and back pain).

The unit of analysis is physician/clinic/year. Each patient is assigned to 1 primary care physician and thus all data will be assigned annually to a physician per clinic. The HSs data will be extracted by numerical ID codes given for each HS.

In this initial phase of the research, we set out to identify MPV patterns by describing and comparing variation between HSs. Our first objective is to identify the main component of variation, that is, which variation is larger, between physicians or within the physician over time (interphysician vs. intraphysician variation). Additionally, we sought to identify which HSs have the highest variation and investigate their common characteristics.

The interphysician variation is the difference in utilization rates between physicians for a given HS, and the intraphysician or within-physician variation is the difference in practice pattern of an individual physician over the years. The difference between the 2 is illustrated in Figure 1. Most research to date examines variation in practice behavior between health care providers at one point in time, not within-physician variation over time. By delineating the 2 (between vs within), we can determine which accounts for a greater proportion of MPV (ie, relative importance of between- and within-physician variation) and what factors predict each. This information is germane to the policy makers, helping them determine where to direct efforts to reduce MPV. For instance, if within-physician variation accounts for a significant proportion of MPV, resources and efforts should be directed at fostering consistent physician clinical behavior over time. However, if between-physician variation is greater, efforts should be directed to assist overusing and underusing physicians to accommodate their utilization patterns to the appropriate level.

After describing the MPV patterns in primary care, we will then assess its determinants. The primary objective of the second phase is to estimate the extent to which each determinant explains the variation and to deduce the overall unexplained variance (Figure 2). In this stage we will collect a wide range of variables related to the domains, calculate the adjusted variance, and the proportional change in the variance (PCV) [117].

In this phase we will visit the clinics and ask the physicians to fill 5 short, validated research questionnaires measuring risk aversion, stress from uncertainty, fear from malpractice, empathy, and burnout [74,118]. We assume that these behavior characteristics are substantially stable [119,120] as well as the practice habits [121]. For instance, burnout [122] or stress from uncertainty is not a temporary emotion, but rather stable, as both are incurred by the properties of the specialty and the physician’s capabilities and characteristics which tend to be fixed [122]. However, to be aligned with the most accurate and updated behaviors and practice habits, we chose in phase 3 to include physicians who worked also during 2017 (approximately 180) as we started to interview them in 2018. Furthermore, we will exclude physicians who worked starting from 2017, as the within-physician variation cannot be assessed for them. To increase the response rate, we plan to conduct face-to-face sessions during which the physicians will be asked to complete the questionnaires. We expect to achieve a response rate of more than 75%.

The study was approved by the Soroka University Medical Center Institutional Ethics Committee (0063-14-SOR). The consent to participate was written as part of the questionnaires.

The data sets that will be used or analyzed during this study will be available following local Ethics Committee approval.

This study will allow us to approach the challenge of the targeted MPV reduction policies by answering a number of questions: What is the degree of the variation and what services have higher variation rates? Which variation is larger: between the physicians or within the physicians over time? What part of the variation cannot be explained by the patient case mix, clinic characteristics, or professional characteristics of the physician? Can physician personal behavioral characteristics explain part of the variation?

Our current study focuses on the variation in the practice patterns, yet we cannot infer the clinical appropriateness of the HSs used. In more general terms we will be not able to deduce what physician is practicing “good” medicine—the one who sends patients to a lot of tests or the one who sends few. Yet, because MPVs have been previously shown to be associated with poorer health outcomes [14,17-19,21], we believe that focusing on the measurement and dissection of the variation itself can contribute to the development of the approaches to reduce the MPV.

Another major limitation of the variation research in medicine is that acceptable values for variation are not defined. Therefore, we will be able only to have a relative comparison identifying factors associated with a higher variation (eg, physician characteristics, specific services).

Furthermore, in some circumstances the utilization of the HSs we are assessing can be considered not discretionary; for example, emergency department visits due to ECG presenting ST elevations in the primary care clinic, vitamin B₁₂ laboratory tests for macrocytic anemia, or chest X-rays for chest injuries. However, the inclusion of these events decreases the variation and thus results in the bias toward zero (null hypothesis).

Moreover, our study precludes system-level factors such as resource constraints, process, workflow issues, funding, services accessibility. This is because we chose to focus on the physician level, not the system level. Therefore, we are only including patients and physicians belonging to 1 health care network (Clalit Health Services), thereby controlling for some of the system-level variation. However, it is possible that the variation between health care networks in the public health system in Israel is limited as the law controls their services’ provision and accessibility, and thus reduces disparities in care. These health care networks are not-for-profit and are prohibited by law from denying any Israeli resident a membership. Yet, between regions in Israel there is a variation in the mentioned factors and therefore our findings can be generalized onto other regions and countries only after accounting for the patterns of this region.

At the final stage of our research we aim to identify behavioral characteristics associated with a higher variation by applying validated questionnaires. Physicians’ attitudes as assessed by the abstract questionnaires may not fully represent their action in real-life clinical practice. Future research should aim at developing more reliable tools for assessing behavioral components of physicians practice.

The conventional risk of questionnaire-based research is a low response rate. To address this concern, we schedule face-to-face sessions with each one of the physicians enrolled into the study.

We expect to analyze a total of 3 million patient years and 6.5 million test utilizations across 16 diverse HSs, referred by 250 physicians in 170 clinics, over 15 years of practice. The size and comprehensiveness of the data will provide a good reassurance for the robustness and generalizability of the research program.

Focusing on the physician personal behavioral characteristics as a major contributing factor to the variation is essential, because it may allow us to identify what are the likely sources of unwarranted variation that can be redressed. We believe that most of the variation explained by patient or clinic characteristics is generally reasonable (eg, greater use by ill and older patients) [20]. However, variation stemming from the physician personal behavior characteristics might be unwarranted and can be reduced without negatively affecting patient care.

Based on the results of our study, we aim to propose a list of potential targets for focused behavioral intervention. The research of behavioral interventions designed for physicians is limited, and describes only a handful of strategies. The most common approaches focusing on changing the practice habits include clinical decision support, shared decision making, pay-for-performance, and insurer restrictions [133]. We believe that identifying new targets for such an intervention during the digital health era can potentially lead to a decrease in the unwarranted variation in the medical practice and thus to the improvement of health outcomes, reduction of disparities in care, and cost savings.