The overall characteristics of the publications included in the review are presented in Table 1. As indicated in this table, the majority of publications are recently published (since 1995). Most publications report empirical analysis of demand for physician services in the USA (labeled as North America in Table 1), although a number of demand studies from European, African and Asian countries are also included in the review. Cross-country comparisons are lacking. The majority of the publications report data collected in household settings. Health care organizations are rarely used as a research setting.

As indicated in Table 1, more than half of all publications directly aim to assess the impact of patient payments on the demand for out-patient physician services. The rest of the publications state that their direct aim is either to model demand for or use of out-patient physician services in general, or to analyze the impact of health insurance on the use of these services. Modeling the choice of physicians is seldom stated as a direct research aim.

A summary of the data collection process reported in the 46 publications is provided in Table 2. The table suggests that the vast majority of publications use data collected in cross-sectional surveys and only few publications report experimental designs, such as controlled or natural experiments. In about half of the publications, sample selection is based on specific population characteristics because authors use existing datasets (e.g., from national surveys) and extract the sample according to research interests in specific population group. Nine publications specify that they used a random sample selection and another ten publications use stratified or cluster random selection.

Table 2 also suggests that the sample size is usually large, i.e., more than 1,000 or even more than 10,000 respondents. However, the response rate is rarely reported. Only a few publications provide information about the response rate and analyze/discuss potential biases due to a selective non-response rate. When the publications report own data collection, the most common methods applied are either interviews or questionnaires.

The characteristics of the data analysis and demand modeling reported in the selected publications are presented in Table 3. Almost all publications included in the review are based on a revealed preferences approach (i.e., data about past consumer behavior). Only one publication reports a stated preferences approach based on statements of consumers about their hypothetical (future) behavior.

More than half of the publications take the probability of visiting a physician or the number of visits to physician as a dependent variable. In all publications, the size of patient payments is included either as an independent variable (in case of a flat-rate co-payment fee) or indirectly as a level of deductibles and/or co-insurance, since this was one of the relevance criteria for the selection of publications for this review. In total 19 publications report models where other (indirect) payments (e.g., travelling and waiting time) are also taken into account. Consumer demographic and socio-economic features, as well as health status are included as independent (explanatory) variables in the majority of the publications. One third of the publications incorporate provider characteristics, such as quality and location.

As indicated in Table 3, most publications use a type of non-linear regression for data analysis (e.g., a type of logit, probit, tobit or count data regression). Linear (i.e., ordinary least-square) regression is relatively less often used. In some cases, two-part or hurdle models are employed, with the first stage formed by the decision to see a physician (e.g., by probit or logit regression) while in the second stage, the number of visits to the physician is analyzed (e.g., by least square regression or count data specification). About a quarter of all publications report simulation methods as an analytical technique. Most publications report other statistical results, such as descriptive statistics, as well as goodness-of-fit statistics.

The main findings of the empirical demand analyses reported in the 46 publications as well as the discussion of their reliability, validity and generalizability are summarized in Table 4. More than half of the publications find that the impact of price (i.e., the patient payment level) on the quantity of out-patient physician services demanded is statistically significant. The price elasticity is reported in two thirds of the publications. Its most usual value is less than 0.50, i.e., relatively low price elasticity. However, 3 publications that report demand analyses for developing countries, indicate higher price elasticity: between 0.50 and 1, and/or higher than 1.

More than half of the publications report statistically significant determinants of the price elasticity of demand for out-patient physician services. Consumer income is overall the most important factor influencing the price elasticity, followed by characteristics of the provider (such as quality and location) and the availability of alternative health care services. One publication reports that the nonmonetary price factors are also significant determinants of price elasticity. Some publications report that price elasticity depends on the price range, i.e., it is higher at the lower price levels and vice versa.

In addition to price, several other factors emerge as important determinants of the demand for out-patient physician services in most or several publications. These include consumers’ socio-economic and demographic characteristics (e.g., age, gender, education, income), time-related payments by consumers, quality and location of service provision, and the price of alternative health care services.

At the bottom of Table 4, the publications are classified according to the validity, reliability and generalizability of the results. A study is considered reliable if the methods of data collection and analysis are well defined and potentially repeatable, and when multiple models or multiple samples produce comparable results. Based on this definition, 19 publications among all publications clearly state the methods of data collection and analysis used, and thus, present a reliable demand analysis. The reliability of the analyses in the rest of the publications is uncertain. Measures of validity included the level of consistence of the obtained results with the stated study hypothesis (theoretical validity), convergence with the results reported in other publications (convergent validity) and indicated attempts to address predictive power of the results (predictive validity). Based on these criteria, 37% of the publications report evidence that confirms certain aspects of validity, namely theoretical, convergent or predictive validity of the findings. Nevertheless, predictive validity is a relatively infrequent subject of discussion. The generalizability of the results is clear in 15% of the publications.

The prevailing approach in modeling demand for out-patient physician services is the revealed preferences approach using data on past consumer behavior. This is probably because the majority of publications reviewed, present analyses based on existing datasets where stated preference data are not available. Utilization of existing datasets is comparatively inexpensive and less time-consuming, but the datasets might lack relevant information (e.g., on social and health characteristics of the study population) that limits the validity of the results. Another consequence of using existing datasets is that it is impossible to estimate response rates and analyze the reasons for non-response together with its potential bias on the interpretation of the results. Additionally, ex-post evaluation of consumer behavior does not provide exact information about the motivation of this behavior and therefore, it does not allow tracking all demand determinants [59]. Thus, causal relations between patient payment policies and the behavioral responses of consumer are difficult to establish.

The stated preference approach to the analysis of demand for out-patient physician services could help to overcome these pitfalls but so far, is rarely applied. The techniques for collecting stated preference data (e.g., discrete-choice experiment and contingent valuation) are relatively new and they are often criticized for their potentially low predictive validity [60–62]. This could explain why we found only one application of the stated preference approach for modeling demand for out-patient physician services. However, the stated preference approach has an important strength. It allows experimentation with policy designs (including patient payment policy) without the need to implement or change this policy. In some instances, actual experimentation with different policy designs might be unethical and socially unacceptable (e.g., experimentation with expansion and reduction of fee magnitudes), or might be impossible if policy is still not implemented. Therefore, it is not surprising that we found only few publications that report experimental designs for data collection.

Even when actual experimentation is possible, data on actual consumer behavior might be interwoven with problems related to co-variation between service attributes (e.g., price and quality) as well as problems related to self-selection and generalization [63]. Individuals who decide for a certain type of behavior are not the random sample of the entire population. Although there are methods to account and correct for the self-selection bias, none of them is completely satisfactory and some of them lead to outcomes that are decidedly disputable [63]. These limitations to the actual experimentation with policy design necessitate the application of stated preference techniques in empirical demand modeling. Furthermore, exploration of mixed approaches, i.e., the combination of stated and revealed preferences methods, to compare actual consumer behavior with statements about this behavior (see [64]) is of particular importance for demand modeling. For example, a mixed approach may be used where a population survey that contains stated preferences questions is followed by a validation experimental stage where a part of the population is included.

The majority of empirical models of demand for out-patient physician services reviewed in this paper, account for relevant determinants of demand, such as socio-demographic characteristics of consumers, direct and indirect (e.g., time-related) payments by consumers, some providers characteristics (such as quality and location), and the price of alternative health care services. However, we did not find studies that investigate in detail the possible influence of provider behavior on consumer demand for out-patient physician services. In particular, supply-side factors relevant to the evaluation of patient payment polices (such as the impact of provider reimbursement schemes and the involvement of providers in the collection of patient payments) are not included in the empirical demand models reviewed. Some publications (e.g., [52]) discuss the possibility of supplier-induced demand as a result of the provider reimbursement scheme applied. Nevertheless, the empirical models of demand presented in these and other publications are based on data from a single country where most supply-side factors lack variability and cannot be analyzed.

In addition to this, not all consumer and patient payment characteristics relevant to the evaluation of patient payment policies are considered in the empirical models of demand reviewed. For example, some consumer-specific factors (such as attitudes, experience and culture), which effect on demand is broadly recognized [6,7], are overall lacking in these demand models. Furthermore, informal patient payments are not considered in any of the publications reviewed. The inclusion of these additional consumer and patient payment characteristics in demand analyses can be crucial because cultural and attitudinal differences among consumers, as well as the level of informal patient payments, may cause different reactions to changes in the patient payment schemes [65]. While from a theoretical point of view these might be important determinants of demand, such data are usually not readily available in existing datasets (as discussed above) or are difficult to collect in case of informal payments [8].

Although the analysis of the empirical results reported in the publications that we reviewed, is not the primary objective of this paper, the overview of these results (see Table 4) suggests some relevant discussion points, namely with regard to price elasticity and its determinants, and other non-price determinants of demand.

Overall, the size of patient payments emerges as an important determinant of demand for out-patient physician services. However, there is a great variety in modeling approaches and inconsistencies of findings regarding the impact of price. In general, the price-elasticity of demand for out-patient physician services is higher in case of lower income, higher patient payments and when alternative health care services exist. Some publications indicate that price significantly influences the quantity of services demanded, while others find this influence statistically insignificant. Moreover, the comparison of the results shows that the values of the price-elasticity reported are rather controversial ranging from very high (higher than 1) to very low (lower than 0.10). This could be explained by the considerable differences in settings, study populations and analytical methods used in the demand analyses. Empirical demand analyses in other areas of the health care sector (namely hospital services) also indicate that the analytical method could be an essential factor when analyzing price-elasticity [36]. Moreover, only few publications report high price elasticity (higher than 0.5) and they are based on data from low-income countries, which may explain the difference with price-elasticity reported in other studies. It should be also acknowledged that not all publications included in our review, address the problem of endogeneity, and prices endogeneity in particular [50]. It may be that insurance coverage, major price changes as well as unobserved preferences, for example due to unmeasured attitudes, experience and culture, influence both demand and prices, and cause the inconsistency in price-elasticity estimates mentioned above. To be able to correct for this, an instrumental variable procedure and generalized method of moments could be used [66–69]. Given the above considerations, it is difficult to conclude whether the price impact in case of out-patient physician services is unambiguous and whether the variations in price-elasticity reported in different publication are genuine. This creates uncertainty regarding the relevance of the existing empirical micro-level models of demand for out-patient physician services for the evaluation of patient payment policies.

Despite the methodological limitations, the existing empirical evidence on price-elasticity of demand for out-patient physician services and its determinants (primarily income), has an important implication. They indicate that the patient payments might have certain potential to influence consumers’ decision whether to use these services, especially among low-income groups. If the demand for out-patient physician services in a given country or in a specific (income) group, is characterized by a high price elasticity, the implementation of patient payments may have a considerable effect on overall efficiency and equity in the provision of such services. For this reason, an adequate demand analysis is essential for the appropriate design of patient payments and the exemption mechanism that should accompany them [1,47,48].

Other possible determinants of demand for out-patient physician services in the publications reviewed include consumer features, time-related payments by consumers, the price of alternative health care services and quality of services provided. Although not all publications confirm the significance of all factors, their potential effects could also be relevant to policy making. For example, when implementing patient payments, health policy makers should take into account non-monetary costs for the population (e.g., waiting and travelling time) when using health care services. In case these costs are high, consumers could be discouraged from seeking out-patient physician services even when provided free-of-charge at the point of utilization [44,45]. At the same time, the location of out-patient physician services in reachable distance could make the introduction or increase of patient payments for these services more acceptable to the population.

The quality of out-patient physician services provided under patient payments also needs to be considered. In particular, monetary payments could have a less discouraging impact when they are followed by quality improvements, such as increased range of services, availability of necessary drugs, and quality of consultations [41]. Quality parameters could have an impact on price responsiveness, although authors [19,42,48] do not completely agree on the magnitude of this impact. The differences in conclusions could be attributed to diverse assumptions about quality parameters and their measurements, as well as to different attitudes within the population studied. Therefore, the importance of various quality parameters for different categories of health care users, as well as the measures of quality from the consumer perspective, need to be taken into account in further research related to the evaluation of patient payment policies.