We used three data sources for the study. First, all reimbursed pharmaceutical products in the National Health Insurance (NHI) database were retrieved as of June 2020. The database, which was provided by the Health Insurance Review and Assessment Service (HIRA), includes information on each product and its manufacturer. The generic and proprietary name, strength, route of administration, and unit price of each product were obtained from the database. Second, all eligible products were identified through the Korea Pharmaceutical Information Service (KPIS) website (25), which provides information on each product, manufacturer, and date of marketing authorization. Finally, KISVALUE (26), an analytic dataset that was created to elucidate manufacturers' financial resources, was used to capture manufacturer characteristics.

We took advantage of a data structure in which pharmaceutical products (level one) were nested within substances (level two). In previous studies, we found that the characteristics of a substance were associated with the price variance of the substance (24). We hypothesized that the characteristics of pharmaceutical products would be associated with price competition. We used a two-level hierarchical model to consider product-related variables at level one and substance-related variables at level two simultaneously. The multilevel model represents price competition for pharmaceutical product i within substance j. At level one, a separate product-level logistic regression model was defined for each substance. At level two, level one coefficients were modeled as functions of level two variables. The term γ₀₀ indicates the average log-odds of price competition across substances, whereas u_0j presents substance-specific intercepts. Numeric variables were group-centered at level one and were grand-mean centered at level two. Categorical variables were not treated as centered.

Level one model

Level two model

PC: price competition.

x1 and x2 are level one predictors.

a1, a2, and a3 are level two predictors.

We applied a null model (model 0), a model with level one variables (model 1), a model with level one and level two variables (model 2), and a model with interaction effects between variables at level one and level two (model 3) to predict price competition. The null model was estimated to investigate whether differences in price competition could be found at the product level or at the substance level. Chi-square tests were used to investigate whether each model fitted the data significantly better than the null model. In a similar vein, we used AIC and BIC to check the goodness of fit of the models. Data management and analysis were performed using R statistical software (version 4.1.3). Particularly, the “lme4” and “lmerTest” package in R statistical software was used to conduct a multilevel logistic regression. Statistical significance was defined as a p < 0.05.

The mean price variance among pharmaceutical products nested in the same substance was 11.67% (24). Given this value, price competition was defined as a discount of >10% in the main analysis. We defined price competition in various ways, including as a discount of >7% and >15%, for sensitivity analyses.

Our analysis included 986 substances that were first granted marketing approval between 2000 and 2019. A total of 12,109 pharmaceutical products were identified for the analysis. Table 1 summarizes the descriptive statistics of all variables included in the model. At level one, the products were grouped into price competition and non-competition categories. Price competition was defined in various ways, such as 7, 10, and 15% discount scenarios. Based on the 10% scenario, 1,381 (11%) out of 12,109 products showed evidence of price competition. The proportion of products exhibiting price competition decreased as the reference point increased. For instance, 14 and 7% of products presented price competition in the 7 and 15% scenarios, respectively. In all scenarios, the types of manufacturers and drug lag measured in years were significantly different between the price competition and non-competition groups. In the 10% scenario, 850 (16.5%) out of 5,154 products produced by major domestic manufacturers exhibited price competition. In comparison, 7.7 and 6.9% of products produced by small- or medium-sized domestic manufacturers and foreign manufacturers, respectively, exhibited price competition. In the same scenario, the average drug lag was shorter for products with price competition (6.59 years) than for those without price competition (7.22 years).

At level two, any substance that included at least one product with price competition was counted in the price competition group. Approximately 50, 43, and 31% of substances were defined as price-competitive in the 7, 10, and 15% scenarios, respectively. We summarized the year of first marketing approval, the maximum reimbursement price, and the number of manufacturers among the price competition and non-competition groups. In all scenarios, the year of marketing authorization and the number of manufacturers differed significantly between the two groups. In the 10% scenario, the mean year of first marketing approval was 2008 and 2006 for the non-competition and price competition groups, respectively. The maximum reimbursement price after log transformation was higher for products in the price competition group than in the non-competition group in the 7 and 10% scenarios. The maximum reimbursement price, however, was not significantly different between the two groups in the 15% scenarios.

Table 2 presents odds ratios for price competition (more than 10% discount) in the market. At level one, manufacturers' characteristics were associated with price competition in models 1, 2, and 3. The odds of price competition for products produced by major domestic manufacturers were greater than the odds of price competition for products produced by foreign manufacturers in model 3B (adjusted odds ratio (AOR) = 10.58, p < 0.0001). In the same model, the odds of price competition for products produced by small- or medium-sized domestic manufacturers were greater than the odds of price competition for products produced by foreign manufacturers (AOR = 5.01, p < 0.0001). However, drug lag was not a consistent predictor in models 1, 2, and 3. Drug lag was associated with decreased odds of price competition in model 1 (AOR = 0.95, p = 0.0251). However, the variable was not a significant predictor in models 2A, 2C, 3A, and 3B.

At level two, the year of marketing approval, the maximum reimbursement price was a consistent predictor in models 2 and 3. However, the variables had different effects on price competition. In model 3B, increases in the year of first marketing approval (AOR = 0.93, p < 0.0001) were associated with decreased odds of price competition, while an increase in the maximum reimbursement price increased the odds of price competition (AOR = 1.27, p < 0.0001). We found interactions between the variables of drug lag at level one and maximum reimbursement price at level two in model 3B. The maximum reimbursement price was a significant predictor of price competition (AOR = 1.27, p < 0.0001), and the effects were stronger when drug lag increased (AOR = 1.05, p < 0.0001).

Table 3 summarizes the findings of sensitivity analysis under various scenarios. Given its goodness of fit, we chose model 3B for sensitivity analysis. We used three different discount scenarios (7, 10, and 15% discounts), and found that some results were consistent across all scenarios. The characteristics of manufacturers at level one, the year of marketing approval and maximum reimbursement price at level two, and the interaction between drug lag and maximum reimbursement price were robust in all scenarios.

In this study, we measured price competition at two levels. At level one, only 11% of pharmaceutical products exhibited price competition in the 10% scenario. In contrast, this value increased to 43% in the same scenario when we applied price competition at level two. Note that any substance that included at least one product with price competition was counted in the price competition group. This finding is consistent with the findings of previous studies (20, 24). One study showed that 75% (82 out of 109) of choline alfoscerate drugs in capsule form retained the maximum reimbursement price of the substance (20). In contrast, the mean price variance among pharmaceutical products containing the same substance has been reported to be 11.67% (24).

Some noteworthy variations in price competition were measured at two levels and suggested that “discounted generic” substitution policies could take advantage of price competition at the substance level. However, the scale of the generic market is relatively small in South Korea. In 2018, the retail market share of generic drugs was 55% in South Korea, making it much smaller than the retail market share of generic drugs in other OECD countries (30). The market shares of generic drugs in the United States (81%), Germany (77%), Canada (76%), and the United Kingdom (71%) all exceeded 70% in the same year (30). In contrast, the average penetration of generics in certain markets in South Korea was reported to be 26.63% in one study (31) and 39.35% in another (32). Policy options that enhance “discounted generic” substitution could be broadened or strengthened to maximize the effects of price competition at the substance level (1, 33, 34).

Previous studies described the accumulation of generic drugs in the South Korean market. Notably, Son (21, 24) categorized generic drug manufacturers as either first-movers or latecomers (21). The designation of first-movers refers to manufacturers entering the market within 2.5 years of the date of the first generic entrant, while latecomers describes manufacturers entering the market after 2.5 years. Latecomers continue to enter mature markets that have been populated by many first-movers. In this way, latecomers ultimately account for the largest portion of the generic drug market. The sizes of manufacturers based on their financial resources are closely associated with their status as a first-movers or latecomers. Major domestic manufacturers have been shown to be more likely to be first-movers.

Compared to foreign manufacturers, major and small- or medium-sized domestic manufacturers were more aggressive at initiating price competition. In particular, major domestic manufacturers play an essential role in health systems insofar as they introduce discounted generics. In this study we also found that foreign manufacturers were more likely to retain the maximum reimbursement price than major domestic manufacturers. In South Korea, foreign manufacturers mainly introduce brand-name drugs into the market, and can thus expect brand loyalty from physicians or patients (35). Brand loyalty implies that foreign manufacturers are reluctant to compete on price with generic entrants, which is a common phenomenon in the United States and Canada (36–38).

When it comes to manufacturers entering the market, order of entry affects the market share of products. This order-of-entry effect assumes that the market favors early entrants due to the benefits of preemption and the brand loyalty of customers (39). Order-of-entry effects have been empirically demonstrated in various pharmaceutical markets (40, 41). Late entrants to a market cannot anticipate high market share for their products. Accordingly, they initiate price competition to overcome their weaknesses in the market and to penetrate a market that consists mainly of early entrants (22–24, 29). Aggressive price competition among late entrants is partially associated with the limited number of generic entrants in other markets. Late entrants are reluctant to enter markets or more willing to offer their products at a discounted price. We observed counterintuitive phenomena in South Korea. First, many late entrants chose to enter the market even after the market had matured. In the sample herein, for instance, 127 out of 986 substances (12.8%) were offered by more than 25 manufacturers. Surprisingly, 39 substances (4.0%) were produced by more than 75 manufacturers. Second, drug lag was not a significant factor predicting price competition in our multilevel analysis. Finally, products with increased drug lag were associated with decreased odds of price competition in some scenarios.

How can we explain such counterintuitive phenomena in the South Korean market? Manufacturers anticipate profit when they decide to enter the market. Profit is calculated by deducting total costs, such as material and labor costs and overhead, from total revenue. In the pharmaceutical market, total revenue is determined by unit price and sales volume. In markets that include aggressive price competition among generic entrants, manufacturers mainly anticipate profits from increased sales. However, manufacturers might also anticipate profits from a product with a high unit price if late entrants have not initiated aggressive price competition. In particular, strict price regulation on generics would result in prices clustering around the maximum reimbursement price with little competition (10, 11). Lack of price competition among generic entrants in South Korea is well-documented in the existing literature (21, 42). Indeed, South Korean manufacturers can reasonably anticipate profits due to higher unit prices, even in cases of late market entry.

This study has several limitations. First, we analyzed price competition among pharmaceutical products following the entry of generics into the South Korean markets. Single-source off-patent or off-exclusivity pharmaceutical products were not included in this study. Second, the lack of availability of certain data should be considered when interpreting the results. We were unable to control for the market size of each substance and the market share of each product in the multilevel model. Including these variables in future research may be beneficial for better understanding. Third, we analyzed pharmaceutical price competition in South Korea, where the prevalence of a large number of generic entrants and a lack of price competition have been reported. The findings of this study may not be generalizable to other markets where a limited number of manufacturers with aggressive price competition are the norm.