Dataset

The data used in the analysis were collected by the Institute for Studies of the Medical Profession (LEFO) in the autumn of 2010. A questionnaire was sent by post to a representative panel of 1 543 Norwegian physicians. The respondents are part of a panel which every other year since 1994 has responded to a questionnaire concerning different aspects of how it is to be a medical doctor. The panel is drawn from the membership database of the Norwegian Medical Association (NMA), which in 2012 include a total of 24 718 physicians, constituting 95% of all physicians practising in Norway. In 1993, 2 000 randomly selected physicians were invited to be part of the reference panel, and 1 300 agreed. Since then two groups of younger doctors have been added to the existing panel to compensate for retirements and ensure the panel is representative [21].

The questions analysed in this study represent a subsection of the larger questionnaire. The subsection contained three parts; demographics, price estimates and questions about attitudes to costs. The questions were developed by reviewing existing literature and the cost pattern of Norwegian drug expenses. The questionnaire was tested on and discussed in detail with five physicians before it was sent out.

Analytical variables

The dependent variable in this study is physicians’ knowledge of the prices of drugs. The questionnaire included five pharmaceuticals for which the respondents were asked to estimate the price. The physicians were asked to estimate the price of one year’s consumption of the standard dose of the pharmaceuticals. The price in question was the sum of the out-of-pocket price paid and the state reimbursement. Respondents were also asked not to check the prices when answering the questionnaire. We chose each specific drug on the grounds of total spending or the number of patients treated with it. We selected a combination of both expensive and inexpensive drugs. When asked for an estimate, respondents were given the medical condition to treat, the name of the drug and the standard dosage as recommended in the Norwegian catalogue of drug information, which is published by the Association of the Pharmaceutical Industry. The pharmaceuticals chosen were simvastatin (cholesterolaemia, 20 mg a day), alendronate (Fosamax) (osteoporosis, 10 mg a day), infliximab (Remicade) (rheumatoid arthritis, 240 mg every 8th week), escitalopram (Cipralex) (depression, 10 mg a day), and natalizumab (Tysabri) (multiple sclerosis, 300 mg infusion every 4th week). Estimates within 50% of the actual price were considered accurate, in line with similar studies [7,9,13]. Some studies accept as accurate only estimates within the range of 25%, but a wider range also eliminates variation or errors in obtaining the correct prices. Correct prices were obtained from the catalogue of drug information for pharmaceuticals [22]. A sensitivity analysis was performed to investigate whether the results would be different when the 25% range was used as the dependent variable (Appendix S1 and S2).

The independent variables in the study are demographic, such as gender, age, medical specialisation and occupational position, and the physician’s attitudes towards using the costs to society as a decision factor when treating patients. Senior consultant is used for all specialties to indicate that the physician has more extensive budget responsibilities.

Attitudes towards costs were measured thus: We asked the respondents to rate ten statements on a Likert scale from 1 (totally disagree) to 5 (totally agree). Also a discrete-choice question was part of this section of the questionnaire. In the discrete choice question, the respondents were asked how much extra they were willing to spend on behalf of society in the following scenario: “A patient with osteoporosis needs to take tablets once a week. A similar treatment is injections that must be given twice a year in hospital. Both treatments are covered by the state but the injections cost more per year than the tablets. How much extra do you think society should pay under various conditions? 1. Patient wants the injections option because they want to avoid the trouble of taking the pills every week. Pills cause mild nausea when taken (once a week). 2. In addition to the patient’s wishes and the nausea, assume that the injection is 10% more effective in increasing bone mass than the tablets.” Physicians willing to pay more than the median estimate of EUR405 (rate NOK to EUR0.135, http://www.xe.com) for scenario 2 are coded as “1” and physicians willing to pay less than this amount are coded as “0”.

Attitudinal variables were designed to determine whether the physicians themselves consider societal costs to be a legitimate decision factor. The sentence “I consider the cost to society when deciding whether or not to initiate an intervention” is a general statement. The sentence “I can reduce my referral and prescribing costs without compromising my patients’ health” will give us an indication of whether the physicians themselves are aware that they are actually practising with slack. Whether or not the physicians spend time checking prices will also say something about whether or not they are willing to support their attitudes with action. We are expecting to find that those who agree that they check prices also estimate prices more accurately than those who say that they do not check prices.

Descriptive statistics and logistic regression in SPSS, IBM Statistics 20 were used to analyse the data. The numbers reported from the logistic regression model are odds ratios (ORs). The dependent variables are dichotomised from the questionnaire items stating the price estimates on each of the pharmaceuticals. All physicians who were off by more than 50% were coded as “1” on this variable and physicians who were within 50%, here defined as giving a correct estimate, were coded “0”. One model has been developed for each of the five pharmaceuticals. The independent variables with an asterisk are those used in the basic model. To avoid co-linearity, closely related variables are substituted in subsequent models. This means that there has only been one specialist variable and one attitude measure in each model. A manual forward stepwise selection process was used to fit the models. Results are adjusted for the physicians’ geographic region of residence, but as we only report statistically significant results, this variable is not included in the here. In the regression analysis, the sample was reduced to include only those who work in clinical practice (n = 740). Excluded respondents were physicians in research, hospital or national administration and retirees. We made this exclusion because we wanted information only from clinicians who actually make decisions about resource allocation.

Response rate and descriptive statistics

After two mailings a total of 1 009 physicians responded, giving a response rate of 65%.

Sample demographics are reported in Table 1. The sample differs from the population to some extent. As of September 2012 the Norwegian Medical Association has registered 22 705 physicians working in Norway below the age of 67, which is the most usual age of retirement in Norway. Forty-six percent of them are female, leaving our sample with a lower proportion of female physicians than in the real population.

The group we left out of the subsample is older than the entire sample of responses, which can be explained by the exclusion of retirees.

Key information about the pharmaceuticals and estimates included in the analysis is displayed in Table 2. Average estimates are below the actual price for the two most expensive pharmaceuticals, and exceed the actual price of the three less expensive drugs. Escitalopram (Cipralex) is, on average, more correctly estimated. More than every second physician on the panel deviates by more than 50% from the actual price, and more than 60% deviate from the actual price by more than 25%.

When physicians were asked how often they check the price of a treatment they would like to offer a patient, 82% responded “never or 1-5 times last year”. Thirteen per cent responded “once a month last year”, while 4% responded “once a week or more during the last year”.

Logistic regression analysis

The results of the logistic regression analysis are reported in Tables 3 and 4. All variables except age are dichotomised and given the values 1 or 0. Age is used as an ordinal variable with the four values below 40, between 40 and 49, between 50 and 59 and above 59. Older respondents were more likely to estimate the price of simvastatin accurately. Since age is used as an ordinal variable, the effect refers to every change in category. A physician in the age group 50-59 has a 19% lower relative likelihood of deviating from the actual price than a physician in the age group 40-49.

Sensitivity analysis

A sensitivity analysis was performed to investigate whether the results would differ when the 25% error range was used as the dependent variable. The model was the same as reported in Tables 3 and 4, but all estimates were considered wrong if they differed more than 25% from the correct price. The results of this analysis are reported in Appendix S1 and S2. A sensitivity analysis to test the difference in results between all respondents to the questionnaire (n 1009) and the subset of clinically active physicians (n 740) was also conducted. The main differences between these analyses and the main models in Tables 3 and 4 are explained in the following.

GPs have a lower relative likelihood of deviating from the actual price by more than 50% on the drugs simvastatin and alendronate in the model with all respondents and on simvastatin in the model testing the 25% error rate.

Physicians specialising in internal medicine are more likely to estimate the price of natalizumab correctly, persisting in all three models.

The effect of age on alendronate persists when the error interval is reduced to 25%, meaning older physicians are more likely to estimate the price of alendronate correctly.

In all three models the physicians who agree with the statement “I can reduce my referral and prescribing costs without compromising my patients’ health” err less on the price of alendronate than the physicians who disagree with this statement. When testing the effect determining a 25% deviation from the actual price the respondents who agree are less likely to deviate from the price of infliximab, suggesting that the physicians who know the price of this drug do not agree that they can reduce their costs.

Another result persisting in all models is the higher odds of senior consultants deviating from the actual price of simvastatin.

When we ran the same model on the entire sample we saw some changes in results. The effect of sex on alendronate is no longer present, but is found for the estimates of infliximab. When reducing the error interval to 25% off the correct price, the effect of sex disappears entirely (see Appendix S2).

In addition to these sensitivity analyses the model has been run using the original attitudinal variables with five values each instead of the binary variables. Dichotomising these variables did not appreciably change the effects.

Implications

The results of this study indicate that physicians might not have the required knowledge to allocate resources efficiently. At the same time, the data also show that there is some potential to adapt and learn the prices of the pharmaceuticals most frequently prescribed by each doctor, especially when the physician perceives cost to be a relevant decision factor.

A fundamental question is whether physicians should have an obligation to consider efficient resource allocation in their role as therapists. Milton Weinstein asked in 2001: “Should physicians be gatekeepers of medical resources?” [28], and answers in the negative. Since their primary obligation is to be their patients’ advocate, Weinstein claims, “it is ethically untenable to expect doctors to face this trade-off during each patient encounter; the physician cannot be expected to compromise the wellbeing of the patient in the office in favor of anonymous patients elsewhere” (p 268).

On the other hand, the physician will have to make some form of resource allocation decisions in practice, whether she/he likes it or not. How much time to spend with a patient, should another test be ordered, or another consultation be booked – these are all choices that any doctor will make on the basis of professional discretion. Discretion in medical care is a necessary part of good medical work, thus the question is not whether the physicians should make allocative decisions, but on what grounds such decisions should be made. Clearly, a fair knowledge of the costs of treatment is one of the requirements in order to make ethically justifiable resource allocation decisions.

Lack of easily available information is one potentially important cause of the poor knowledge of costs. In a 1995 survey of Norwegian doctors, 51% said they did not have enough information about the cost of their daily decisions to be able to make good decisions about community resources [29]. If we acknowledge that physicians make choices which impact on how the total health-care resources are allocated between tasks and patients, and as such should be expected to choose cost effective treatments, they must be given the proper information and decision support tools. We see from other studies that physicians do not have good access to cost information, and often they believe that they could prescribe more cost effectively if they had access to better information [11,30,31]. In a recent study, Feldman, Shihab et al. concluded that displaying the prices of diagnostic tests can influence physicians ordering behaviour and lead to a reduction in costs [32].

Methodological issues

Variations in the price of pharmaceuticals could constitute a problem because in this case there is no single true price. In Norway the regional variations are small because a maximum price is set centrally. There is some variation between the prices paid by hospitals, which have a separate pricing system, and the prices paid by others. However, the questions were worded to reduce this potential confusion by emphasising that price should be interpreted to include the payment by the patient, for the pharmaceuticals this is relevant for, and no patient pays for pharmaceuticals when they receive them in hospital.

The validity of the results depends on the representativeness of the sample. One potential problem might be that only those interested in economic issues answered the questions. Since the survey was part of a larger biennial survey, this problem is minimised, yet there is always the general problem that respondents may be more organised and knowledgeable than non-respondents. In this sense the results are conservative estimates of physicians’ knowledge of prices.