1 INTRODUCTION

We study physicians' adoption of innovative treatment programs. Healthcare stake holders such as insurers, government, and sponsors would like to motivate physicians to adopt best practices. However, physician practices vary, and often variations have not been associated with better health outcomes or efficient resource uses (Skinner, 2012). Motivating the adoption of effective protocols may yield higher efficiency.

The best ways to motivate adoption are seldom agreed upon. The extant economics literature has concentrated on financial incentives and competition. The health and medicine literature, however, has exhibited a wider perspective (see, e.g., Gawande, 2010). Paying physicians for adopting certain procedures or treatments, and pay-for-performance contracts are examples of financial mechanisms, whereas information dissemination, peer reinforcement, educational programs, and explicit enforcement of new protocols are examples of nonfinancial mechanisms.

In this paper, we study mechanisms for the adoption of new information system and treatment guides for preventive care. The study setting is the monitoring of Type 2 Diabetes (T2D) in Norway between 2009 and 2014; we use primary care physician claims and register data in this time period. The Norwegian Directorate of Health (2009, 2016) publishes medical guidelines for T2D prevention, diagnostics and treatment. Patients with T2D may benefit from monitoring and prevention from deterioration, so should have a structured annual check-up with their primary care physicians.1 After the annual check-up, physicians submit data to a national quality register, which is maintained by a center called NOKLUS.2

The financial aspect of the Norwegian Directorate guideline works through a fee-for-service module. From 2008, the structured annual check-up has entitled a physician to a payment, in addition to the office visit fee. This additional charge is called Fee 109, which was about NOK 110 in 2012 (equivalent to about US $20 in 2012). Fee 109 has been a national policy. There has also been a regional education program to promote the adoption since 2013. In two west coast counties, Rogaland and Hordaland, diabetes nurses have been installing software, and providing training to physicians. Both financial and educational policies are thought to promote adoptions. Although Fee 109 and the education program were not set up for comparing their relative effectiveness, the selective education implementation can be regarded as a quasi-experiment.

Using Norwegian physician panel data, we study financial and educational mechanisms in multiple steps. First, we use fixed-effect models to study how physician and municipality characteristics affect Fee 109 adoption. We identify a peer effect; those physicians who practice in municipalities with more adopters tend to adopt Fee 109 and use the checkup more often; peer effects show up in the extensive margin. We find that physician competition has a negligible effect on adoption. A separate analysis on 230 physicians who have moved between municipalities supports the identification. Second, to assess education, we use a difference-in-difference regression for the program that started in 2013; we find a strong causal effect on Fee 109 use. The program has a greater effect in municipalities where many physicians already have adopted, again confirming that education programs' adoption effects are contingent on local adoption shares. Having peers who have already adopted strengthens the education program effect. Third, we use two-part models and hazard models to verify robustness of the fixed-effect models.

We are unaware of research that uses national micro data to study primary care medical technology adoption, which we believe is an important issue. First, primary care plays a crucial role in an aging population because of the prevalence of chronic illnesses such as T2D and chronic obstructive pulmonary disease among the elderly. As well, primary care physicians provide the crucial lead in care coordination between generalist and specialists. Second, technology in primary care technology is less sophisticated than technology in hospitals and specialty care. Lowbrow technology and protocols such as surveillance and monitoring have probably escaped researchers' attention, but actually deserve more investigation because they can be cost effective. Third, primary care is decentralized as single or small-group practices, so very different from hospitals. Results on hospital technology adoption cannot be expected to apply to primary care.

The structure of the paper is as follows. The next subsection is a literature review. Section 2 describes the institutional setting of Norwegian health care. Section 3 presents a theory of optimal adoption decisions and derives hypotheses. Data and descriptive statistics are in Section 4. Our main empirical results are in Section 5, where we present fixed-effect estimation to assess peer effects, and difference-in-difference estimation to assess the effectiveness of education programs. Robustness checks are in Section 6, where we present estimation results from two-part models and hazard-rate estimations from survival models. Finally, Section 7 draws some conclusions.

1.1 Literature review

Encouraging physicians to adopt technology and treatment guidelines has received attention in the literature. However, according to Grol and Grimshaw (2003), many physicians are slow to adopt. The economics literature and the medical literature seem to have looked at different perspectives on adoption. Whereas the economics literature focuses on financial incentives and competition, the medical literature focuses on audits, peer reviews and educational programs.

Grol (1992) suggests that physicians' reluctance to adopt stems from competence, attitude, and personal characteristics such as age and training. Indeed, continuing medical education, face-to-face instruction, audit and feedback can encourage adoption. Wensing et al. (1998) find that social influence and management support can improve information transfer, but performance information or ratings do not. Ivers et al. (2012) find that audit and feedback improve professional practice and patient outcomes, although the effect can be small. Our paper fits into this literature. The education program for physicians in specific Norwegian counties served as a quasi-experiment on T2D prevention and monitoring technology adoption.

Economists' recent interest on social network builds a bridge between the medical and economics literature. A recent review, Miraldo et al. (2019), documents the role of peers and networks on technology diffusion: evidence shows that physician characteristics and network collaboration improve information dissemination, especially when best practice is not agreed upon. Molitor (2018) makes use of cardiologists' migration to study the role of practice environment on physician behavior. He finds that physician behavior in the first year after the move changes 0.6–0.8 percentage points for each percentage point change in practice environment. Our finding is in-line with Molitor's: where peers have adopted, physicians tend to adopt more often.

Most of the economics empirical literature on technology adoption is about hospitals and specialty care. Baker (2001) examines the relationship between Health Maintenance Organization (HMO) market share and magnetic resonance imaging (MRI) diffusion. Across markets, higher HMO market shares are associated with slower MRI diffusion, and markedly lower MRI uses. Horwitz et al. (2017) study the adoption of three invasive cardiac services from 1996 to 2014 (diagnostic angiography, percutaneous coronary interventions, and coronary artery bypass grafting). Using proportional hazard models, they find that hospitals are more likely to adopt an invasive service if hospitals within an hour of driving distance also adopt new services. Karaca-Mandic et al. (2017) find that drug-eluting stent diffusion is faster where cardiology practices face more competition. In our study, competition seems to have played little role. First, we have no exogenous competition policy changes in the data period. Second, Fee 109 does not require a huge capital investment, and the market demand has not responded to adoption.

Our study is related to the literature on factors that affect physicians' new drug adoption. Liu and Gupta (2012) use a micro-level diffusion model and find that marketing efforts, patients' requests, and contagion effects of nearby physicians have shown positive adoption influence. In Kremer et al. (2008), a meta-analysis shows significant and positive adoption effects by promotional expenditures, but these are modest and vary across diseases. Blankart and Stargardt (2020) document that German Health Technology Assessment agencies' information dissemination in the form of published favorable assessments leads to quicker adoption as well as favorable negotiation and pricing.

Studies of technology diffusion in primary care is quite scarce. Scott et al. (2009) evaluate the impact of an incentive program in primary care in Australia on diabetic care quality, measured by the probability of ordering an Hemoglobin-A1c blood sugar test. The study finds that the incentive program increases the probability of an HbA1c test by 20 percentage points. Klausen et al. (1992) study the diffusion of dry-chemistry equipment in Norwegian primary care practices. Based on the maximization of future net revenue of a practice, the adoption probability at a certain date should be positively related to incremental income, practice consultations, and dry-chemistry analysis reimbursement, but negatively related to wet-chemistry reimbursement, and dry-chemistry equipment prices. Their empirical work finds support that diffusion is affected by profits. These papers have generally adopted a benefit-cost approach, to which we have subscribed here.

Our setting of T2D monitor recommendation has been in a number of descriptive studies. Using records of patients identified with diabetes mellitus, Claudi et al. (2008) present cross-sectional results from four Norwegian geographical areas. About 90% of the study population had HbA1c tests, and blood pressure and lipids measurements annually. More than 70% of patients with T2D were referred for eye examinations; albumin levels were recorded in 40% of patients. The authors concluded that care quality improved substantially, but potential improvements were possible. Bakke et al. (2017) compare the results in Claudi et al. (2008) with those in a 2014 survey among physicians. They find moderate improvements during the previous decade, which confirms a worldwide trend. Perhaps more pertinent for our work, Bakke et al. (2018) find that performance varies substantially between physicians; physicians with a higher workload tend to order fewer procedures. They conclude that performance of screening procedures was suboptimal overall, and that the use of a structured diabetes form should be mandatory.

What are potential benefits and costs to patients and society from more annual check-ups? The potential health benefits and reduced costs are considerable if check-ups avert chronic decline and complications. For Sweden, Andersson et al. (2020) find substantial economic burden due to T2D complications. Key cost drivers are the macrovascular complications angina pectoris, heart failure and stroke; and the microvascular eye diseases, including retinopathy, kidney disease and neuropathy. Early mortality in working ages contributes to a substantial production loss in addition to the health care costs.

Can systematic check-ups result in fewer late disease complications? In a literature review of the association between GPs recording clinical data and T2D mortality and morbidity, Larun et al. (2016) conclude that current published data provide ambiguous answers, but form use in diabetic patient follow-ups in general practice may lower mortality and morbidity. An important question is whether a regular check-up may associate with recommended procedure uses. Nøkleby et al. (2020) find large variations in T2D patient care delivered by physicians, who performed on average 63.4% of the recommended procedures (with 46% in the lowest quintile and 81% in the highest). The structured follow-up was associated with GPs being in the top three quintiles. In Nøkleby et al. (2021), the authors explore the association between recommended procedure uses and patients' cardiovascular risk. They find that patients treated by the 20% worst-performing physicians have a 75% higher modifiable cardiovascular risk compared with patients of the 20% best-performing physicians. Hence, altogether the potential benefits and averted costs to patients and society from increased use of the annual check-up may be substantial.

In summary, the health economics literature finds that economic incentives have an impact on technology adoption. The evidence has come from hospital and specialty care, but not from primary care. The medical literature finds that education programs, audits and feedback may have positive effects on adoption although magnitudes differ across studies.

2 STUDY SETTING 2.1 Norwegian health services and primary care physicians

Norwegian National Health Service provides health care for more than 5 million residents. Since 2001 each resident has been offered to list with a primary care physician, who provides primary care and serves as a gatekeeper for specialty care. By 2010 over 95% of the population participated in the list system. In 2010, over 95% of more than 4100 Norwegian primary care physicians were private practitioners who contracted with municipalities (For brevity, from now on, the term physician means primary care physician.) The remaining physicians were salaried municipality employees. Physician employees usually work in sparsely populated areas; a fixed salary serves to shield physicians from financial risks due to service demand fluctuations in low-population areas. In the present paper, we consider only private-practice physicians, and all descriptions and analyses apply to them only.

The list system comes with the following financial arrangements for physicians. First, the physician receives a capitation fee from the municipality for each patient in her list; in 2012, this fee was NOK 386 per year, at which time the exchange rate was about US $1 to NOK 6. A physician had, on average, 1200 patients listed in his practice. Second, a physician receives fee-for-service payments, set by the National Insurance Scheme, when health services are provided to patients. Third, the physician also receives a patient copayment at the time of service; the copayment is decided by the Norwegian Parliament as part of the National Insurance Scheme. Each revenue component constitutes about one third of a physician's practice income.

In a calendar year, a patient may switch physicians up to two times, and each year approximately 3% of the patients do switch. Characteristics of patients who switch vary considerably. Patients who are male or older, and who have good health but only basic education tend to stay with their physicians. Switching patients form a kind of market demand. In the supply side, a physician sets the maximum practice list size. A practice may actually be closed when a physician has enough patients. Whether a practice is open or closed is public information, available on the Internet or from the municipality. A physician may have fewer patients than the declared maximum. In the empirical work, a physician is said to experience shortage if the actual list falls short of the stated maximum by more than 100 patients. Patient shortage and not being a specialist in general medicine make it more likely that physicians experience patients switching into or out of their practices (Iversen & Lurås, 2011).

Whereas patients receive general care from physicians, they receive specialty care from specialists, who may be private practitioners or work in public hospitals. Most private specialists contract with Regional Health Authorities, which are responsible for hospitals in their regions. A private specialist receives a practice allowance from a Regional Health Authority, and fee-for-service payments from National Insurance Scheme. Most private specialists are in urban areas, and they provide about one third of all outpatient consultations. In 2012, a patient's copayments for an outpatient visit with a physician and a specialist were about NOK 180 and NOK 307, respectively, but a patient's copayment within a year was capped at NOK 1980 and any excess was paid for by the National Health Insurance.

2.2 Type 2 diabetes and the annual comprehensive check-up

We consider technology adoption for Type 2 diabetes (or T2D) management. Diagnostic and treatment guidelines have been developed in countries with different health systems for this common chronic illness. For instance, in the United States, Kaiser Permanente (2017) presents detailed guidelines for monitoring T2D patients. The monitoring includes glycemic control target, microalbumin assessments and regular retinal and foot screening. Similar guidelines have been worked out by Socialstyrelsen (2018) for Sweden. A recent study in France (Andrade et al., 2018) shows that adherence to four guidelines (quarterly determination of HbA1c, complete lipid profile, microalbuminuria and influenza vaccination) is associated with monitored patients having up to 30% fewer annual hospital admissions.

The Norwegian Directorate of Health (2009, 2016) publishes medical guidelines for diabetes prevention, diagnosis, and treatment. National medical experts work out the guidelines. T2D is included in the guidelines together with Type 1 Diabetes (T1D). The guidelines prescribe that T2D patients should have physician check-ups. For patients with poorly regulated diabetes or complications, physicians and specialists should share care responsibility to coordinate treatment.

From 2006, Norwegian Quality Improvement of Primary Health Care Laboratories (NOKLUS) has initiated a national quality register, The Norwegian Diabetes Register for Adults. The Register was approved by the Norwegian Data Inspectorate in 2005. The goal is to develop a T2D patient database. Medical personnel submit data to the Register on a voluntary basis, subject to patients' written consent. For data submission, physicians have to install computer software that links to patient electronic records. The software also assists the physician with organizing the annual check-up to include all required components. The Register issues annual quality reports to participating medical centers and individual doctors. Hence, the technology we study contains two components. The first is an annual check-up according to medical guidelines. The second is a software, supplemental to patient electronic records, assisting the physician to implement the annual check-up, and transferring data to the national quality register.

Each time a physician uses the annual checkup for T2D patients and submits data to the Register, she can, in addition to the consultation fee, file a fee-for-service claim, the Fee 109, which was about NOK 110 in 2012. To mitigate coding errors, we conservatively define physician adoption by her claiming Fee 109 at least 10 times annually. Despite the recommendation of NOKLUS by the Directorate of Health and the Fee 109 reimbursement, only a minority of physicians have chosen to participate. Accordingly, there has been an interest in identifying participating physicians' characteristics. Furthermore, to encourage participation, since 2013, physicians in Rogaland and Hordaland, two counties on the Norwegian west coast, have been offered assistance. A diabetes nurse would install software for the comprehensive annual check-up and launch data submission to the national register. The assistance also includes an education session to demonstrate the working and the benefit from the checkup software. These counties were chosen because they obtained project funding. Also, one diabetes nurse was already based at the Register in Hordaland, and a physician had both a position at the Register and at Stavanger University Hospital in Rogaland.

Patient associations have often an important role in the education of patients with chronic disease. In Norway, the Norwegian Diabetes Association is an influential patient association with 33,000 members (in 2018). This accounts for about 15% of the patients with T1D and T2D. We do not have access to the distribution of members according to counties. We do know the number of local organizations by counties. Adjusted for population size, Hordaland has an above-median number of local organizations, whereas Rogaland has a below-median number.

The second part of our empirical work studies whether this educational effort has increased adoption. In effect, we regard the efforts for Rogaland and Hordaland as a quasi-experiment.

3 ADOPTION DECISION AND HYPOTHESES

We focus on a physician's decision on the adoption of the technology for monitoring patients who have chronic illnesses, so abstract from other such decisions as practice size, amount of fee-for-service treatments, referrals, etc. We simply posit that the physician's adoption decision is based on a benefit-cost comparison. We then hypothesize how the adoption benefit and cost depend on a physician's personal characteristics and prevailing market conditions.

Consider the adoption decision to be made by a physician in a municipality at a certain point in time. We use a binary variable to represent the adoption decision; takes the value 0 if the physician does not adopt, and the value 1 if she adopts. We use a vector to denote the physician's personal characteristics, and another vector to denote market conditions. We let the function denote benefits, and the function denote costs; these functions may well be indexed by physicians, municipalities, and time, but we will gloss over these indexes for a simpler exposition.3 Benefits and costs can be financial and nonfinancial, and represent discounted values. Adoption may change the patient list and service demand, which, in turn, change revenues and job satisfaction; likewise, service and time cost may change due to adoption. Due to list size and service demand uncertainty, the benefit and cost functions are to be regarded as the expected benefits and expected costs that arise from the adoption decision.

We assume that adoption results in a benefit increment: . The new monitoring technology would be valuable to patients with a chronic illness, so may yield financial or altruistic benefits. We naturally assume that adoption is costly: . A physician's adoption decision is now described by the choice of that maximizes . Equivalently, a physician adopts whenever the benefit increment is higher than the cost increment: .

Obviously, a physician's adoption decision depends on her personal characteristics, those represented in . In the empirical study, we have information about such physician characteristics as (i) age, (ii) medical specialty, (iii) the percentage of patients with a chronic illness in the practice, and (iv) other factors. The decision may also depend on market conditions, those represented in , such as (i) number or percentage of other physicians who have adopted, (ii) population density and access to specialty care, and (iii) competition, which we take to be the numbers of other physicians who accept new patients.

How do a physician's personal characteristics affect adoption benefits and costs? The physician likely enjoys higher adoption benefits (i) when she is younger, so has a longer career horizon, (ii) when she is a specialist in primary care, and (iii) when her practice has more patients who suffer from T2D. In a symmetric way, the physician likely has a higher adoption cost (i) when she is older, (ii) when she does not specialize in primary care, and (iii) when few patients in the practice suffer from T2D.

How do current adoption levels and market conditions affect adoption decisions? Within a market, (i) when more physicians have adopted the technology, a peer effect may develop, so learning may be easier and conforming with the norm may be preferred. We predict that when a market has more physicians who already have adopted the technology, it is more likely that a physician adopts. Also, within a market, (ii) when consumers have better access to specialty care, the demand for monitoring may be less, so the adoption benefit is reduced. Finally, (iii) competition as measured by the number of open practices, either in nominal or in per-capita terms, may have an ambiguous effect on benefits. Through word of mouth chronically ill patients may learn about a physician using the structured checkup; this practice style may be perceived as a quality increase. However, adoption may not be so attractive to those who are not chronically ill. Overall, it is ambiguous how competition intensity may change the benefits from adoption.

4 DATA AND DESCRIPTIVES

Data for this paper come from two sources. The first one is primary care physicians' claims to the National Insurance Scheme. This database is called KUHR. The second source is the regular primary care physician register that contains information on physicians' characteristics, as well as identifies physicians' patient lists. Data are aggregated to the physician level, and supplemented with patients' residential municipality data.

Patients who have chronic diseases are identified from KUHR. Claims data contained diagnoses at each contact. We identified patients with T2D from the diagnosis code T90 in International Classification of Primary Care. Patients with T2D were those who received the diagnosis code T90 in at least one consultation between 2006 and 2009. The two data sources cover the 6 years between 2009 and 2014, and are merged at the level of the individual patient's physician.

We define a physician as an adopter if she makes use of Fee 109 at least 10 times in 1 year. The 10-time use criterion is to avoid counting as adopters those physicians who have filed claims by mistake. Our definition also yields adoption-percentage figures that are consistent with those in NOKLUS. We drop from the data set those physicians who have had less than 10 T2D patients. Figure 1 shows that the proportion of physicians using Fee 109 has increased from 5% in 2009 to just above 10% in 2014. These percentages are approximately equal to the proportions of physicians who submitted data to The Norwegian Diabetes Register for Adults in 2009 and 2014.

Time trend of physicians adopting Fee 109

Table 1 presents 2009 descriptive of adopters and non-adopters. Where appropriate, in all tables standard errors are in parentheses, and significance levels are indicated by the usual convention of * for , ** for , and *** for . On average adopters have more patients with T2D than non-adopters; adopters' patients tend to have more comorbidities. Adopters are more likely to be specialists in primary care, but have the same average age as non-adopters. Physicians report to the National Insurance Administration the maximum number of patients allowed in their practices (the maximum list or Maxlist). Adopters have longer patient and maximum lists, and are less likely to experience patient shortage than non-adopters.

TABLE 1. Physician descriptive statistics in 2009 Variable definition Non-adopters Adopters Mean difference Mean SD Mean SD Physician level

#T2D

Number patients with T2D

48 28 min 10 66 33 min 26 − max 307 max 305

Proportion T2D

Proportion patients with T2D

0.04 0.02 min 0.01 0.05 0.02 min 0.02 −0. max 0.17 max 0.25

#comorbidities

Number comorbidities

18 12 min 1 25 19 min 4 − max 143 max 185

Specialist

Primary care specialist

0.60 0.49 min 0 0.74 0.44 min 0 −0. max 1 max 1

Age

Physician age

48 10 min 26 48 10 min 28 0 max 80 max 67

Maxlist

Maximum list size

1192 441 min 0 1245 350 min 0 −53 max 2500 max 2400

List

Actual list size

1219 380 min 157 1317 326 min 541 − max 2720 max 2396

Shortage

List < (Maxlist – 100)

0.20 0.40 min 0 0.15 0.36 min 0 0. max 1 max 1 Municipality level

Total_listed

Listed patients (in 1000)

111 190 min 0.1 88 153 min 0.7 23 max 612 max 612

#openlists

Number open lists

30 56 min 0 21 43 min 0 max 183 max 183

#open_per_cap

0.40 0.41 min 0.00 0.40 0.31 min 0.00 0 max 8.26 max 1.55

#adopters

Total number of adopters

5.6 8.4 min 0 7.1 9.2 min 1 −1. max 33 max 33

Access private specialist

Private specialist access index

0.68 1.11 min −1.54 0.46 1.09 min −1.33 0. max 2.21 max 2.21

Access hospital

Hospital access index

1.72 3.66 min −2.09 1.19 2.79 min −1.26 0. max 11.78 max 11.78 Observations 3669 201

On average adopters are located in smaller municipalities than non-adopters. Adopters' smaller municipalities have fewer open lists than non-adopters’ municipalities. However, adopters and non-adopters seem to have the same percentage of practices with open lists per 100,000 inhabitants. Table 1 also describes the number of physicians in the municipalities who have adopted the comprehensive annual check-up (#adopters). Adoption seems to be more likely when there are more adopters in the previous years in the same municipality. The variable “Access private specialists” is an indicator for access to private specialists; it weighs the number of specialists by patients' travel time from the listing physician's practice municipality. The variable “Access hospital” is a similar variable for physician specialists located in hospitals. From Table 1 access to private and hospital specialists is better in non-adopters’ municipalities than adopters'.

Table 2 shows descriptive in 2014. The number of adopters has increased from 201 in 2009 to 445 in 2014. Nevertheless, differences between adopter and non-adopter profiles, both at the physician level and at the municipality levels, have not changed much between 2009 and 2014.

TABLE 2. Physician descriptive statistics in 2014 Variable definition Non-adopters Adopters Mean difference Mean SD Mean SD Physician level #T2D 37 22 min 10 51 23 min 15 − max 252 max 227 Proportion T2D 0.03 0.02 min 0.00 0.04 0.02 min 0.01 −0. max 0.14 max 0.19 #comorbidities 14 9 min 0 19 12 min 2 − max 106 max 130 Specialist 0.60 0.49 min 0 0.70 0.46 min 0 −0. max 1