SURVEY

We had two sets of questionnaires mailed to GPs in the spring of 1999, one set to all GPs (n = 739) who had HPRT in their office laboratories, and a different set to a random sample (n = 717) of GPs who did not. The response rate after one reminder was 57 percent in both groups. The questionnaire contained a clinical vignette, describing a 30-year-old woman with dyspepsia, which was used to assess the clinical reasoning and decisions made by GPs. The vignette should be fairly familiar to the GP, and in fact, with some modifications, this case history was based on journal notes of a real patient. Minor modifications were made in collaboration with several clinicians (GPs and gastroenterologists) and a microbiology specialist. It was an important element of the vignette that additional tests should not be necessary.

The GPs were asked to state:

• – the pre-test probability that Mrs. Hansen's dyspepsia was caused by H. pylori,
• – whether or not they would order HPRT or the serological test in this situation,
• – which actions they would take based on the history or on the history in addition to the HPRT result (not dealt with in this paper).

In addition, we obtained information on the characteristics of all the GPs.

HYPOTHESES

Table 1 gives an overview of the variables included. The fixed cost of using a test consists of resources related to evaluating the information about the test in addition to the expenses of performing test in the office laboratory (the staff and the lab utilities needed). The variable cost is the cost of the actual test kit. Hence, with a considerable fixed cost, the average cost of having and using a test is a declining function of the number of tests. We therefore predict that the probability of having and using HPRT will increase with the number of consultations per week. Likewise, the average cost of having HPRT will decrease with the number of GPs in the practice, and we therefore assume that group practices have a higher probability of having HPRT. As a consequence of more intercollegiate exchange of information in group practices, doctors are more aware of its limitations and, thus, use it differently than solo practitioners.

General practices in urban areas face competition for patients, and one way of competing is to give quicker service to the patients. Thus, we assume that GPs in the cities have a higher probability of having the HPRT and using it in the office lab than GPs in rural or semi-urban areas.

Primary health care in Norway is the responsibility of the municipalities. In principle, GPs are either in private practice being paid fee-for-service, or they are on a fixed salary as employees. In our study, most were fee-for-service, and their income derives from three sources: a fixed grant from the municipality (or a per capita fee for list doctors), patient copayment, and reimbursement from the National Health Insurance according to a negotiated tariff (approximately one third each).

In contrast to GPs on a fixed salary, GPs in private practice in general may increase their income by using lab tests. The reimbursement for the HPRT was NOK 90, and the cost of the test was around NOK 77, so the net surplus is quite small. However, when the test was introduced on the market, the reimbursement was NOK 175, thus promoting the inclusion of this test in the repertoire. The GPs who submit a serological test to an approved laboratory were reimbursed with only NOK 25 in 1999. We assume here that the expenses connected with performing the test in the office laboratory are covered by the reimbursement, but direct economic incentives for using this test are weak, and non-existent for GPs on a fixed salary. Still, use of this in-office laboratory test may be less time-consuming than sending away a serological test, because the doctor will get the result the same day, with less need for a follow-up appointment. Thus, the GPs perceived availability of his services may be of importance but is probably independent of the remuneration system.

Upper endoscopy is the preferred diagnostic procedure for detecting peptic ulcer and is done in hospitals. Especially in younger patients, HPRT is an alternative to upper endoscopy. Norway is a scarcely populated country, and many inhabitants live hours away from the nearest hospital. The inconvenience to the patient increases with increasing waiting and traveling times and is lessened if an office lab is used.

The pre-test probability is the GP's assumption that the patient in our case history had an H. pylori infection before the HPRT is administered. We assume that each GP generally has his own threshold value at or above which he considers the patients' probability of having an H. pylori–related dyspepsia sufficiently great to take some kind of action The exception is when the probability is very high or low; then the GP feels confident of the diagnosis, making the use of the HPRT superfluous. We test this hypothesis by including a squared pre-test variable that we assume is negative, because the probability of using lab is increasing after a lower threshold value and decreasing after a higher threshold value.

Having HPRT in the general practice may indicate a higher general awareness of peptic ulcer disease caused by H. pylori infection; hence, this category of GPs would be more apt to use the lab test.

We asked the GPs whether they need more information about the use of the HPRT. GPs not having the HPRT may need information to evaluate whether they should have the test. For GPs having the test, it's important to be aware of how to use the test and its characteristics (sensitivity and specificity). Accordingly, we cannot a priori say whether how the need for information influences the probability of having the test.

DATA

With the exception of the GPs pre-test probability, we have only used the background variables in the questionnaire (ref. Table 1). We use the same data for having HPRT (level A) and using lab (level B), except that at level B we include the variable “pre-test probability,” and use “has HPRT” as an independent variable.

A total of 210 of 425 GPs with HPRT decided to use this test, but only 100 GPs of 410 GPs without HPRT decided to use the serological test. We excluded a few observations when the GP clearly had misunderstood the question, or groups of GPs with deviant characteristics: GPs on internship in general practice, those over 67 years of age, those with working hours more than 60 or less than 10 per week, number of consultations more than 160 or less than 10 per week, and waiting time more than 26 weeks (for referral).

After exclusion, we had 201 GPs with HPRT and 84 GPs without HPRT who requested this analysis (9 and 16 doctors, respectively, excluded). Overall, 14.4 percent of GPs included were publicly employed, but more detailed data show that only 6 percent of GPs having HPRT and only 8 percent of GPs using the test in the lab were publicly employed.

The decision at level A is common to all the GPs in the practice. In our data, we had 432 general practices; and of them, 233 were solo practices. To estimate the effect of characteristics by the general practice, we grouped the data per general practice, and divided by the number of general practitioners in the practice responding to the questionnaire to get the average or share.

We compared the characteristics age, male, and type of payment of our sample of GPs with the total population of GPs (from a registry kept by the Norwegian Medical Association). We found that our sample had the same mean values regarding age, had a higher share of men (77 percent versus 73.6 percent), and only had half the share of GPs on fixed salary (14.4 percent versus 28 percent).

EMPIRICAL MODELS

The theoretical framework is based on discrete choice analyses from Ben-Akiva and Lerman (2) and Greene (6). We want to establish models for predicting the probability of a general practice choosing to have the HPRT (level A) and the probability of a GP using the test in the office lab (level B), as a function of the characteristics of the general practice at level A and as a function of the characteristics of the GP at level B. At both levels, the GP has two possibilities that are mutually exclusive and we use binary models.

All the GPs evaluates the same patient vignette—so the focus here is on the GP's own objectives and preferences, knowledge, experience, and uncertainty, not knowing the patient's preferences in this clinical situation.

We suppose that the GPs have preferences for the different alternatives that can be represented by a utility function. We suppose that the utility of an alternative depends on the GP's income, or cost (income to the GP) associated with the alternative. Furthermore, we also assume that the welfare of the patient is associated with the GP's utility. There is a trade-off between time and quality. The HPRT is time saving, but the serological test is better (higher sensitivity and specificity).

The utility of the HPRT depends on the type of patient in question, how important it is to get the answer during the same consultation, and the GP's knowledge about the use of the HPRT. We assume that the income is included in the GP's utility function. For GPs with fee-for-service, we assume that the considerable fee when the test was introduced in 1996 (NOK 175) and to some degree the GP's perception of the development of the fee of the HPRT will be of importance.

The GPs' choice setting can be compared with choices between lotteries, because of the uncertainty of the initial health status of the patient and the laboratory analysis. The uncertainty of the laboratory analysis occurs because the HPRT measures antibodies to the H. pylori bacteria and not the disease as such and because healthy carriers of the H. pylori bacteria exist. In younger people, the prevalence of the bacteria may be as high as 15 percent.

The GP may also have unstable preferences, meaning that he or she may have problems in evaluating the expected utility of the different alternatives. This possibility means that he may make different responses when faced with the same choices (bounded rationality). The degree of bounded rationality may vary in apparently identical situations because colleagues, medical journals, and experience gained from treating other patients have a continuous influence on the GP. Furthermore, there will be variation in the choices that cannot be explained by the variables available to us, especially those stemming from the patient–doctor interaction.

We assume that the utility U_ij is stochastic and can be interpreted as the expected utility for the general practice_i at level A and for the GP_i at level B, given alternative j, j = 1,2. Our binary logit models can be formulated as follows:

where Y_i* is unobservable and Privpr_i is a dummy for type of payment: Privpr_i =1 for GPs in private practice, and Privpr_i =0 for GPs on fixed salary. X_i is a vector of the other independent variables and ε_i is a random variable that is supposed to count for unobserved variables of the GP that affects his preferences.

If every independent variable did not correlate with the random variable, ε_i, the derivation of the probability for choosing between alternatives 1 and 2 would be straightforward. But other studies have shown that the variable “Privpr” depends on characteristics of the GP. Grytten and Sørensen (8) have shown that there is a self-selection, that GPs on fixed salary give more priority to family and leisure. If unobserved characteristics of the GP influence the variable “Privpr,” it becomes an endogenous variable and will correlate with the error term.

We want to study the factors that influence the GP's type of payment. GPs with a high activity (number of consultations and working hours) may prefer payment by activity. The selection effect will be that GPs with high activity prefer contracts with reimbursement. To study the effect of selection, we will apply a particular instrumental variable approach. To this end, we will estimate a model for g(Z_i) = E(Privpr_i[mid ]Z_i), which we will use as an instrumental variable for the variable Privpr_i, where Z_i is a vector of the independent variables (working hours, X_i [vector of the other independent variables]). One problem is that the error terms in the models with and without the instrumental variable are different, because the instrumental variable is not perfect, and we cannot directly compare the coefficients.

ESTIMATION RESULTS

The results are estimated using LIMDEP (7). Table 2 shows the values, the p values, and odds ratios of the coefficients in the model. For the variable location, urban area is used as reference. We have included results from the model both without and with an instrumental variable (Table 2).

Availability of the HPRT in General Practice (Level A)

Here, we examine the importance of the characteristics of the general practice (with one or more GPs) for the probability of having HPRT by estimating a binary logit model where Has HPRT (Y =1)/Does not have HPRT (Y =0) is the dependent variable.

We had 420 observations, and the log Likelihood function for the model with the instrumental variable is -278.7616.

The only difference between the models with and without the instrumental variable is that the remuneration variable is only significant without the instrumental variable. This finding means that the effect of the variable “Privpr” without the instrumental variable was due to the selection of GPs choosing fee for service.

In estimating the instrumental variable (g[Z_i]) the dependent variable is private practice and the independent variables are the variables we believe influence the GP's choice of payment: male, age, education, location of the practice, and number of consultations. McFaddens R² for the instrumental variable is 33.37.

Table 2 shows that the following factors increase the probability of having HPRT on the lab repertoire: whether the GPs in the general practice want information about the use of HPRT and whether the GPs in the general practice have many consultations.

Use of the H. pylori Analysis (Level B)

In this section, we study the effect of different characteristics on the probability of using the test in the lab by estimating a binary logit model with “Uses lab (Y =1)/Does not use lab (Y =0)” as the dependent variable.

The main difference between the models with and without the instrumental variable is that the remuneration variable is only significant without the instrumental variable. This finding means that the effect of the variable “Privpr” without the instrumental variable was due to the selection of GPs choosing type of payment. Also, traveling time (for gastroscopy) just falls short of significance when the instrumental variable is included, probably because location of the practice is included in this variable.

We had 663 observations and the value of the log likelihood function was -356.7273 for the model with the instrumental variable, which in this analysis is calculated for the individual GP. McFaddens R² for the instrumental variable is 28.6.

Table 2 shows that the following factors have a significant effect on the probability of using the test in the office lab: GPs in group practices use office lab tests less than solo practitioners, this lab test is used less by rural GPs, the availability of the rapid test increases test use substantially, and in-office lab use increases with pre-test probability.

CONTACT INFORMATION

Siri Fauli, MBE, PhD student (siri.fauli@legeforeningen.no), Institute of Health Management and Health Economics, University of Oslo, P.O. Box 1089 Blindern, NO-0317 Oslo, Norway; Adviser, Department of Professional Affairs, The Norwegian Medical Association, P.O. Box 1152 Sentrum, N-0107 Oslo, Norway

Geir Thue, MD, PhD, Professor (geir.thue@isf.uib.no), Department of Public Health and Primary Health Care, The Norwegian Center for Quality Improvement of Laboratory Services in Primary Care (NOKLUS), University of Bergen, UNIFOB AS, Boks 6165 Postterminalen, 5892 Bergen, Norway