- Split View
-
Views
-
Cite
Cite
Sylvia Park, Stephen B Soumerai, Alyce S Adams, Jonathan A Finkelstein, Sunmee Jang, Dennis Ross-Degnan, Antibiotic use following a Korean national policy to prohibit medication dispensing by physicians, Health Policy and Planning, Volume 20, Issue 5, September 2005, Pages 302–309, https://doi.org/10.1093/heapol/czi033
- Share Icon Share
Abstract
This study investigated whether a Korean national policy prohibiting doctors from dispensing drugs as of 2000 selectively reduced inappropriate antibiotic prescribing in viral illness compared with bacterial illness.
We assessed the proportions of episodes prescribed an antibiotic and the number of different antibiotics prescribed for patients with viral and bacterial illness episodes before and after the policy. The nationally representative sample consisted of 50 999 episodes (18 656 viral and 7758 bacterial pre-policy, 16 736 viral and 7849 bacterial post-policy) from 1372 primary care clinics. We used generalized estimating equations to investigate changes in antibiotic prescribing after the policy, and multiple linear regression to determine provider factors associated with reductions in inappropriate antibiotic prescribing for viral illness.
After the dispensing restriction, antibiotic prescribing declined substantially for patients with viral illness (from 80.8 to 72.8%, relative risk (RR) = 0.89, [95% confidence interval: 0.86, 0.91], p<0.001), and only minimally for patients with bacterial illness (from 91.6 to 89.7%, RR = 0.98, [0.97, 0.99], p = 0.017). Reductions in antibiotic prescribing were significantly larger (RR = 0.90, [0.87, 0.93], p<0.001) for patients with viral illness.
The number of different antibiotics prescribed per episode also decreased significantly after the policy, but there were no significant differences in these reductions between viral and bacterial illness. The dispensing restriction also reduced prescribing of non-antibiotic drugs, with no difference by diagnosis. Provider factors found to be associated with reduced inappropriate antibiotic prescribing were younger age and practice location in an urban area.
Prohibiting doctors from dispensing drugs reduced prescribing overall, both of antibiotics and other drugs, and selectively reduced inappropriate antibiotic prescribing in viral illness.
Introduction
Drug prescribing is an important determinant of health care expenditures and overall quality of care. There has been increasing concern about physicians who both prescribe and dispense medications with economic incentives (Lavizzo-Mourey and Eisenberg 1990; Morton-Jones and Pringle 1993; Brown et al. 1995; Baines et al. 1996; Nizami et al. 1996; Trap and Hansen 2002a,b; Trap et al. 2002; Chou et al. 2003). Although the majority of physicians in industrialized countries do not dispense drugs, some industrialized countries and many Asian countries permit this dual role (Lavizzo-Mourey and Eisenberg 1990; Chou et al. 2003), and the number of dispensing doctors is increasing in several countries (Trap et al. 2002).
When physicians have a financial incentive to dispense medications, they are likely to prescribe more drugs. In previous research, dispensing doctors, in comparison with non-dispensing doctors, were found to prescribe greater numbers of drugs (Brown et al. 1995; Nizami et al. 1996; Trap et al. 2002), more antibiotics and injections (Trap et al. 2002), and to have higher prescribing costs (Morton-Jones and Pringle 1993; Baines et al. 1996; Chou et al. 2003). Dispensing doctors have also been associated with less appropriate prescribing (Trap et al. 2002). Due to practical barriers, research on physician dispensing has mostly compared prescribing behaviours between dispensing and non-dispensing doctors cross-sectionally (Morton-Jones and Pringle 1993; Trap and Hansen 2002a,b; Trap et al. 2002). No studies have examined how quality of prescribing changes when physician dispensing is disallowed.
In July 2000, a new Korean government policy prohibited physicians from dispensing and pharmacists from prescribing drugs by law. Prior to the policy, all office-based physicians dispensed drugs in their clinics; the national health insurance system reimbursed them for dispensed drugs with prices pre-determined differently for each drug, allowing physicians to profit on mark-up over drug purchase costs. After the policy, physicians could no longer dispense drugs, which removed a potential profit incentive. It was expected that the policy would change physicians' prescribing behaviours. This provided a natural experiment to investigate the impact of a policy prohibiting medication dispensing by physicians in the same population.
Antibiotics are among the most commonly used therapeutic agents. In Korea, antibiotics were used at a rate of 33.2 DDD/1000 inhabitants/day in 1997, much higher than the average in OECD countries of 21.3 DDD/1000 inhabitants/day (Lee et al. 2000). It has been reported that a considerable proportion of antibiotic use in Korea is inappropriate or unnecessary (Park and Moon 1998). Excessive use of antibiotics is thought to be an important contributor to growing worldwide antibiotic resistance (Arason et al. 1996; Arnold et al. 1996; Wang et al. 1998; Chen et al. 1999). Korea has much higher resistance rates than other countries, with 86% of Streptococcus pneumoniae isolates resistant to penicillin, compared with rates of 46% and 34% in Singapore and the US, respectively (Doern et al. 2001; Lee et al. 2001).
The purpose of this study is to evaluate the impact on the quantity and quality of physicians' prescribing of the new policy in Korea that prohibits physicians from dispensing drugs. Specifically, we investigate changes in antibiotic prescribing after the policy and whether the reduction in antibiotic prescribing was greater for cases of viral disease, in which antibiotic prescribing was likely inappropriate, than for bacterial disease, in which antibiotic prescribing could be appropriate.
Methods
Data sources and research design
Data were extracted from the Korean national health insurance claims data system, which covers more than 95% of the population and contracts with all clinics and hospitals in Korea. Clinics submit monthly claims for reimbursement. Each claim represents an episode of care for each patient, and contains diagnoses coded according to the International Classification of Diseases, Tenth Revision (ICD-10), demographic characteristics, clinic identification, number of visits, and all health services and drugs prescribed for the episode that month.
We used data for the months of January 2000 and January 2001, which were 6 months before and after the policy intervention, respectively. Between these periods, we know of no other policy changes or events which might have influenced drug utilization.
Based on previously published studies (Finkelstein et al. 2000, 2003), we identified two primary diagnosis groups of interest. The viral illness group included: bronchiolitis (ICD-10 code: J219), common cold (J00), and upper respiratory tract infection (J06, J060, J068, J069). The bacterial illness group included: bacterial pneumonia (J13 to J15, J150 to J154, J158, J159, J168, J180, J189), suppurative otitis media (H66, H660 to H664, H669), streptococcal sore throat (J020), tonsillitis (J039), sinusitis (J01, J010 to J013, J018, J019, J32, J320 to J323, J328, J329), urinary tract infection (N300, N309, N390), and skin and soft tissue infections (L023, L024, L028, L029, L03, L030 to L033, L038, L039, L04, L05, L050, L059, L08, L088, L089).
From a database containing a random sample of 20% of claims from a randomly selected 10% of clinics operating in both time periods (n = 1476), we included all claims that had one of the above diagnosis codes as a primary diagnosis (n = 124 944). We excluded records with a secondary diagnosis (59% of all records) in order to eliminate the possibility that the secondary diagnosis might justify antibiotic prescribing. Thus 26 414 eligible illness episodes before the policy and 24 585 episodes after the policy from 1372 clinics were included in the analysis. Only one episode per patient was included during the study period.
Prescription variables
The primary outcome studied was whether an antibiotic was prescribed for the episode of illness. For episodes in which an antibiotic was prescribed, a secondary outcome was the number of different antibiotics (different generic names or formulations) prescribed for the episode of illness.
To evaluate if changes observed were specific to antibiotics, we also examined the number of different non-antibiotic drugs prescribed as well as gastrointestinal drug prescribing (including antacids, antiemetics, H2 receptor antagonists, laxatives) for these episodes. Gastrointestinal drugs were chosen because they were frequently used during the illness episodes studied, but were not essential for treating the primary infection.
Statistical analysis
Characteristics of eligible illness episodes
For each diagnosis group before and after the policy change, we summarized the patient characteristics (i.e. gender and age) and clinic characteristics (i.e. urban/rural location of clinic, practice size according to the number of patients in the sample database [≤150 patients, 151–250 patients, ≥251 patients], and type of practice [solo/group]) associated with the identified illness episodes. For solo practice clinics, we also characterized physicians' gender and age.
Policy effects on prescribing
The dependent variable (Yi) in our primary model was a binomial term indicating whether or not an antibiotic was prescribed during the illness episode. Clinic was specified as the cluster effect. Key explanatory variables included a policy variable (coded 0/1) indicating whether the episode in question occurred before or after the new policy, and a variable indicating diagnosis group (viral, coded 1 or bacterial, coded 0). A third explanatory variable, specified as the interaction between the policy and diagnosis group variables, measured our primary variable of interest, that is, the differential effect of the policy on antibiotic prescribing in the viral diagnosis group compared with the bacterial diagnosis group. We also included in the final model any characteristics of patients and providers which were associated (p<0.05) with antibiotic prescribing in univariate analyses and which remained significant in the multivariable model. In the GEE model, we specified a binomial distribution with a log link function in order to produce estimates of the relative risk (eβ) of antibiotic prescribing. Since antibiotic prescribing is prevalent in our study, the relative risk is preferable to the odds ratio, which overstates effects when the event rate is high (Robbins et al. 2002).
We used similar models to evaluate changes in the number of different antibiotics, the number of different non-antibiotic drugs, and the probability of gastrointestinal drug prescribing. In the analyses of number of antibiotic/non-antibiotic drugs, which were counted variables, we applied the Poisson distribution of GEE with log link function to produce estimates of the ratio (eβ) of the numbers of antibiotic/non-antibiotic drugs.
Provider characteristics associated with decrease in inappropriate antibiotic prescribing
At the clinic level, we next investigated provider characteristics associated with inappropriate antibiotic prescribing (i.e. prescribing an antibiotic for a viral illness episode) at baseline and with decreases in inappropriate antibiotic prescribing following the policy change. Clinics were included in these analyses only if they had at least 10 viral illness cases in both periods, so that we could calculate outcomes at the clinic level.
For our baseline analyses, the dependent variables were the adjusted clinic-specific proportions of viral illness episodes prescribed an antibiotic and the average number of different antibiotics per viral illness episode before the policy. We then analyzed changes in these measures after the policy was implemented. Independent variables included location, practice size and practice type. When we limited our analysis to solo practice clinics, which constituted 90% of all clinics, additional independent variables included physician's gender and age. Only variables with significant effects (p<0.05) in the multiple regression analysis were included in the final models.
Results
Characteristics of episodes
Characteristics of episodes by diagnosis group before and after the dispensing restriction policy are given in Table 1. Of the 50 999 illness episodes, 46% involved male patients and 53% involved children (i.e. age ≤18). Most episodes (89%) were treated in clinics located in urban areas, and 43% of episodes were from clinics with more than 250 patients in the sample. Episodes from solo practice clinics, which accounted for 90% of all episodes, were predominantly treated by male physicians. Physicians in solo practice had an average age of 45.3 years.
Characteristics . | Percentage of illness episodes . | . | . | . | |||
---|---|---|---|---|---|---|---|
. | Bacterial diagnosis . | . | Viral diagnosis . | . | |||
. | Before . | After . | Before . | After . | |||
Total | 100.0 | 100.0 | 100.0 | 100.0 | |||
(n) | (7758) | (7849) | (18 656) | (16 736) | |||
Gender of patient | |||||||
Male | 47.6 | 46.2 | 46.5 | 45.9 | |||
Female | 52.4 | 53.8 | 53.5 | 54.1 | |||
Age of patient (years) | |||||||
≤2 | 12.7 | 10.3 | 21.1 | 22.8 | |||
3–18 | 41.3 | 39.1 | 30.9 | 32.5 | |||
19–64 | 42.0 | 47.3 | 42.7 | 40.5 | |||
≥65 | 4.0 | 3.3 | 5.3 | 4.2 | |||
Location of clinic | |||||||
Urban | 91.5 | 92.0 | 87.5 | 89.4 | |||
Rural | 8.5 | 8.0 | 12.5 | 10.6 | |||
Practice size | |||||||
≤150 patients | 20.8 | 24.0 | 21.7 | 24.0 | |||
151–250 patients | 30.2 | 32.8 | 36.2 | 34.0 | |||
≥251 patients | 49.0 | 43.2 | 42.1 | 42.0 | |||
Type of practice | |||||||
Solo | 90.1 | 90.1 | 88.3 | 89.3 | |||
Group | 9.9 | 9.9 | 11.7 | 10.7 | |||
Gender of physician | |||||||
Male | 90.2 | 89.4 | 92.4 | 90.2 | |||
Female | 9.8 | 10.6 | 7.6 | 9.8 | |||
Age of physician (years) | |||||||
≤39 | 34.9 | 34.0 | 28.2 | 29.5 | |||
40–49 | 50.1 | 48.9 | 45.6 | 47.3 | |||
≥50 | 15.0 | 17.1 | 26.2 | 23.2 |
Characteristics . | Percentage of illness episodes . | . | . | . | |||
---|---|---|---|---|---|---|---|
. | Bacterial diagnosis . | . | Viral diagnosis . | . | |||
. | Before . | After . | Before . | After . | |||
Total | 100.0 | 100.0 | 100.0 | 100.0 | |||
(n) | (7758) | (7849) | (18 656) | (16 736) | |||
Gender of patient | |||||||
Male | 47.6 | 46.2 | 46.5 | 45.9 | |||
Female | 52.4 | 53.8 | 53.5 | 54.1 | |||
Age of patient (years) | |||||||
≤2 | 12.7 | 10.3 | 21.1 | 22.8 | |||
3–18 | 41.3 | 39.1 | 30.9 | 32.5 | |||
19–64 | 42.0 | 47.3 | 42.7 | 40.5 | |||
≥65 | 4.0 | 3.3 | 5.3 | 4.2 | |||
Location of clinic | |||||||
Urban | 91.5 | 92.0 | 87.5 | 89.4 | |||
Rural | 8.5 | 8.0 | 12.5 | 10.6 | |||
Practice size | |||||||
≤150 patients | 20.8 | 24.0 | 21.7 | 24.0 | |||
151–250 patients | 30.2 | 32.8 | 36.2 | 34.0 | |||
≥251 patients | 49.0 | 43.2 | 42.1 | 42.0 | |||
Type of practice | |||||||
Solo | 90.1 | 90.1 | 88.3 | 89.3 | |||
Group | 9.9 | 9.9 | 11.7 | 10.7 | |||
Gender of physician | |||||||
Male | 90.2 | 89.4 | 92.4 | 90.2 | |||
Female | 9.8 | 10.6 | 7.6 | 9.8 | |||
Age of physician (years) | |||||||
≤39 | 34.9 | 34.0 | 28.2 | 29.5 | |||
40–49 | 50.1 | 48.9 | 45.6 | 47.3 | |||
≥50 | 15.0 | 17.1 | 26.2 | 23.2 |
Characteristics . | Percentage of illness episodes . | . | . | . | |||
---|---|---|---|---|---|---|---|
. | Bacterial diagnosis . | . | Viral diagnosis . | . | |||
. | Before . | After . | Before . | After . | |||
Total | 100.0 | 100.0 | 100.0 | 100.0 | |||
(n) | (7758) | (7849) | (18 656) | (16 736) | |||
Gender of patient | |||||||
Male | 47.6 | 46.2 | 46.5 | 45.9 | |||
Female | 52.4 | 53.8 | 53.5 | 54.1 | |||
Age of patient (years) | |||||||
≤2 | 12.7 | 10.3 | 21.1 | 22.8 | |||
3–18 | 41.3 | 39.1 | 30.9 | 32.5 | |||
19–64 | 42.0 | 47.3 | 42.7 | 40.5 | |||
≥65 | 4.0 | 3.3 | 5.3 | 4.2 | |||
Location of clinic | |||||||
Urban | 91.5 | 92.0 | 87.5 | 89.4 | |||
Rural | 8.5 | 8.0 | 12.5 | 10.6 | |||
Practice size | |||||||
≤150 patients | 20.8 | 24.0 | 21.7 | 24.0 | |||
151–250 patients | 30.2 | 32.8 | 36.2 | 34.0 | |||
≥251 patients | 49.0 | 43.2 | 42.1 | 42.0 | |||
Type of practice | |||||||
Solo | 90.1 | 90.1 | 88.3 | 89.3 | |||
Group | 9.9 | 9.9 | 11.7 | 10.7 | |||
Gender of physician | |||||||
Male | 90.2 | 89.4 | 92.4 | 90.2 | |||
Female | 9.8 | 10.6 | 7.6 | 9.8 | |||
Age of physician (years) | |||||||
≤39 | 34.9 | 34.0 | 28.2 | 29.5 | |||
40–49 | 50.1 | 48.9 | 45.6 | 47.3 | |||
≥50 | 15.0 | 17.1 | 26.2 | 23.2 |
Characteristics . | Percentage of illness episodes . | . | . | . | |||
---|---|---|---|---|---|---|---|
. | Bacterial diagnosis . | . | Viral diagnosis . | . | |||
. | Before . | After . | Before . | After . | |||
Total | 100.0 | 100.0 | 100.0 | 100.0 | |||
(n) | (7758) | (7849) | (18 656) | (16 736) | |||
Gender of patient | |||||||
Male | 47.6 | 46.2 | 46.5 | 45.9 | |||
Female | 52.4 | 53.8 | 53.5 | 54.1 | |||
Age of patient (years) | |||||||
≤2 | 12.7 | 10.3 | 21.1 | 22.8 | |||
3–18 | 41.3 | 39.1 | 30.9 | 32.5 | |||
19–64 | 42.0 | 47.3 | 42.7 | 40.5 | |||
≥65 | 4.0 | 3.3 | 5.3 | 4.2 | |||
Location of clinic | |||||||
Urban | 91.5 | 92.0 | 87.5 | 89.4 | |||
Rural | 8.5 | 8.0 | 12.5 | 10.6 | |||
Practice size | |||||||
≤150 patients | 20.8 | 24.0 | 21.7 | 24.0 | |||
151–250 patients | 30.2 | 32.8 | 36.2 | 34.0 | |||
≥251 patients | 49.0 | 43.2 | 42.1 | 42.0 | |||
Type of practice | |||||||
Solo | 90.1 | 90.1 | 88.3 | 89.3 | |||
Group | 9.9 | 9.9 | 11.7 | 10.7 | |||
Gender of physician | |||||||
Male | 90.2 | 89.4 | 92.4 | 90.2 | |||
Female | 9.8 | 10.6 | 7.6 | 9.8 | |||
Age of physician (years) | |||||||
≤39 | 34.9 | 34.0 | 28.2 | 29.5 | |||
40–49 | 50.1 | 48.9 | 45.6 | 47.3 | |||
≥50 | 15.0 | 17.1 | 26.2 | 23.2 |
Policy effects on antibiotic prescribing
Table 2 presents the unadjusted proportion of episodes treated with antibiotic/gastrointestinal drugs and the average number of antibiotics/non-antibiotic drugs prescribed by diagnosis group before and after the policy. It also presents the GEE results for the effects of the dispensing restriction policy on prescribing in the viral and bacterial diagnosis groups, and differential effect on viral group compared with the bacterial group, after adjusting for clustering and characteristics of cases.
Outcome . | Diagnosis group . | Before the policy . | After the policy . | Policy effectsa . | . | ||
---|---|---|---|---|---|---|---|
Relative risk (95% CI) | p value | ||||||
Proportion of episodes | Bacterial | 91.5% | 89.6% | 0.98 (0.97–0.99)b | 0.017 | ||
prescribed an antibiotic | Viral | 80.3% | 73.2% | 0.89 (0.86–0.91)c | <0.001 | ||
Differential policy effect in viral groupd | 0.90 (0.87–0.93)e | <0.001 | |||||
Proportion of episodes | Bacterial | 84.2% | 79.1% | 0.96 (0.93–0.98)b | 0.003 | ||
prescribed | Viral | 79.4% | 72.8% | 0.92 (0.90–0.95)c | <0.001 | ||
gastrointestinal drugs | Differential policy effect in viral groupd | 0.97 (0.93–1.01)e | 0.124 | ||||
Ratio (95% CI) | |||||||
Number of different | Bacterial | 1.7 | 1.6 | 0.94 (0.92–0.96)b | <0.001 | ||
antibiotics per episode | Viral | 1.5 | 1.4 | 0.92 (0.90–0.95)c | <0.001 | ||
Differential policy effect in viral groupd | 0.99 (0.96–1.01)e | 0.357 | |||||
Number of different | Bacterial | 4.8 | 4.5 | 0.92 (0.90–0.95)b | <0.001 | ||
non-antibiotic drugs | Viral | 3.9 | 3.5 | 0.93 (0.91–0.95)c | <0.001 | ||
per episode | Differential policy effect in viral groupd | 1.01 (0.97–1.05)e | 0.745 |
Outcome . | Diagnosis group . | Before the policy . | After the policy . | Policy effectsa . | . | ||
---|---|---|---|---|---|---|---|
Relative risk (95% CI) | p value | ||||||
Proportion of episodes | Bacterial | 91.5% | 89.6% | 0.98 (0.97–0.99)b | 0.017 | ||
prescribed an antibiotic | Viral | 80.3% | 73.2% | 0.89 (0.86–0.91)c | <0.001 | ||
Differential policy effect in viral groupd | 0.90 (0.87–0.93)e | <0.001 | |||||
Proportion of episodes | Bacterial | 84.2% | 79.1% | 0.96 (0.93–0.98)b | 0.003 | ||
prescribed | Viral | 79.4% | 72.8% | 0.92 (0.90–0.95)c | <0.001 | ||
gastrointestinal drugs | Differential policy effect in viral groupd | 0.97 (0.93–1.01)e | 0.124 | ||||
Ratio (95% CI) | |||||||
Number of different | Bacterial | 1.7 | 1.6 | 0.94 (0.92–0.96)b | <0.001 | ||
antibiotics per episode | Viral | 1.5 | 1.4 | 0.92 (0.90–0.95)c | <0.001 | ||
Differential policy effect in viral groupd | 0.99 (0.96–1.01)e | 0.357 | |||||
Number of different | Bacterial | 4.8 | 4.5 | 0.92 (0.90–0.95)b | <0.001 | ||
non-antibiotic drugs | Viral | 3.9 | 3.5 | 0.93 (0.91–0.95)c | <0.001 | ||
per episode | Differential policy effect in viral groupd | 1.01 (0.97–1.05)e | 0.745 |
aPolicy effects on drug prescribing adjusted for clustering and patient and provider characteristics from GEE.
beβ1 from GEE.
ceβ1 + β3 from GEE.
dDifferential policy effect in viral group compared with bacterial group.
eeβ3 from GEE.
Outcome . | Diagnosis group . | Before the policy . | After the policy . | Policy effectsa . | . | ||
---|---|---|---|---|---|---|---|
Relative risk (95% CI) | p value | ||||||
Proportion of episodes | Bacterial | 91.5% | 89.6% | 0.98 (0.97–0.99)b | 0.017 | ||
prescribed an antibiotic | Viral | 80.3% | 73.2% | 0.89 (0.86–0.91)c | <0.001 | ||
Differential policy effect in viral groupd | 0.90 (0.87–0.93)e | <0.001 | |||||
Proportion of episodes | Bacterial | 84.2% | 79.1% | 0.96 (0.93–0.98)b | 0.003 | ||
prescribed | Viral | 79.4% | 72.8% | 0.92 (0.90–0.95)c | <0.001 | ||
gastrointestinal drugs | Differential policy effect in viral groupd | 0.97 (0.93–1.01)e | 0.124 | ||||
Ratio (95% CI) | |||||||
Number of different | Bacterial | 1.7 | 1.6 | 0.94 (0.92–0.96)b | <0.001 | ||
antibiotics per episode | Viral | 1.5 | 1.4 | 0.92 (0.90–0.95)c | <0.001 | ||
Differential policy effect in viral groupd | 0.99 (0.96–1.01)e | 0.357 | |||||
Number of different | Bacterial | 4.8 | 4.5 | 0.92 (0.90–0.95)b | <0.001 | ||
non-antibiotic drugs | Viral | 3.9 | 3.5 | 0.93 (0.91–0.95)c | <0.001 | ||
per episode | Differential policy effect in viral groupd | 1.01 (0.97–1.05)e | 0.745 |
Outcome . | Diagnosis group . | Before the policy . | After the policy . | Policy effectsa . | . | ||
---|---|---|---|---|---|---|---|
Relative risk (95% CI) | p value | ||||||
Proportion of episodes | Bacterial | 91.5% | 89.6% | 0.98 (0.97–0.99)b | 0.017 | ||
prescribed an antibiotic | Viral | 80.3% | 73.2% | 0.89 (0.86–0.91)c | <0.001 | ||
Differential policy effect in viral groupd | 0.90 (0.87–0.93)e | <0.001 | |||||
Proportion of episodes | Bacterial | 84.2% | 79.1% | 0.96 (0.93–0.98)b | 0.003 | ||
prescribed | Viral | 79.4% | 72.8% | 0.92 (0.90–0.95)c | <0.001 | ||
gastrointestinal drugs | Differential policy effect in viral groupd | 0.97 (0.93–1.01)e | 0.124 | ||||
Ratio (95% CI) | |||||||
Number of different | Bacterial | 1.7 | 1.6 | 0.94 (0.92–0.96)b | <0.001 | ||
antibiotics per episode | Viral | 1.5 | 1.4 | 0.92 (0.90–0.95)c | <0.001 | ||
Differential policy effect in viral groupd | 0.99 (0.96–1.01)e | 0.357 | |||||
Number of different | Bacterial | 4.8 | 4.5 | 0.92 (0.90–0.95)b | <0.001 | ||
non-antibiotic drugs | Viral | 3.9 | 3.5 | 0.93 (0.91–0.95)c | <0.001 | ||
per episode | Differential policy effect in viral groupd | 1.01 (0.97–1.05)e | 0.745 |
aPolicy effects on drug prescribing adjusted for clustering and patient and provider characteristics from GEE.
beβ1 from GEE.
ceβ1 + β3 from GEE.
dDifferential policy effect in viral group compared with bacterial group.
eeβ3 from GEE.
The unadjusted proportion of bacterial illness episodes prescribed an antibiotic was 92% before the policy, and 90% after the policy. The relative risk (RR) derived from GEE was 0.98 (95% confidence interval: 0.97, 0.99; p = 0.017). In viral illness, 81% and 73% of episodes were prescribed an antibiotic at baseline and after the policy, respectively (RR = 0.89 [0.86, 0.91], p<0.001). Antibiotic reductions were significantly larger (RR = 0.90 [0.87, 0.93], p<0.001) for episodes of viral illness compared with bacterial illness episodes for which antibiotic prescribing may be justifiable.
Antibiotic polypharmacy was quite widespread. At baseline, the average number of antibiotics per episode was 1.7 in episodes of bacterial illness and 1.5 in episodes of viral illness; these were reduced after the policy to 1.6 (ratio = 0.94 [0.92, 0.96], p<0.001) and 1.4 antibiotics (ratio = 0.92 [0.90, 0.95], p<0.001) per episode, respectively. The reduction in the number of antibiotics was not different between the two diagnosis groups.
Policy effects on non-antibiotic drug prescribing
The proportion of gastrointestinal drug prescribing also dropped significantly after introduction of the new policy, in both viral and bacterial diagnosis groups, but the differences were not statistically significant. Similarly, the number of different non-antibiotic drugs prescribed decreased after the policy in both diagnosis groups, with no significant difference between groups (Table 2).
Provider factors associated with decrease in inappropriate antibiotic prescribing
We examined the predictors of antibiotic use and of reductions in use after the policy change at the clinic level for viral illness episodes. Table 3 presents the summary of multiple regression analyses of the factors predictive of inappropriate antibiotic prescribing. The analyses included 435 and 351 clinics at baseline and after the policy, respectively. Average numbers of episodes per clinic before and after the policy were 38.3 (sd 28.5) and 36.4 (sd 24.7), respectively.
Outcome . | Provider characteristics . | Beta . | 95% CI . | p value . |
---|---|---|---|---|
Proportion of antibiotic prescribing at baseline | Practice type (group)a | −0.14 | (−0.23, −0.05) | 0.002 |
Changes in number of antibiotics per episode | Age (≤39)b | −0.13 | (−0.24, −0.02) | 0.024 |
Age (40–49)b | −0.16 | (−0.26, −0.06) | 0.002 | |
Changes in number of antibiotics per episode | Practice location (urban)c | −0.14 | (−0.27, −0.01) | 0.030 |
Outcome . | Provider characteristics . | Beta . | 95% CI . | p value . |
---|---|---|---|---|
Proportion of antibiotic prescribing at baseline | Practice type (group)a | −0.14 | (−0.23, −0.05) | 0.002 |
Changes in number of antibiotics per episode | Age (≤39)b | −0.13 | (−0.24, −0.02) | 0.024 |
Age (40–49)b | −0.16 | (−0.26, −0.06) | 0.002 | |
Changes in number of antibiotics per episode | Practice location (urban)c | −0.14 | (−0.27, −0.01) | 0.030 |
areference: solo practice.
breference: age 50 and older.
creference: rural area.
Outcome . | Provider characteristics . | Beta . | 95% CI . | p value . |
---|---|---|---|---|
Proportion of antibiotic prescribing at baseline | Practice type (group)a | −0.14 | (−0.23, −0.05) | 0.002 |
Changes in number of antibiotics per episode | Age (≤39)b | −0.13 | (−0.24, −0.02) | 0.024 |
Age (40–49)b | −0.16 | (−0.26, −0.06) | 0.002 | |
Changes in number of antibiotics per episode | Practice location (urban)c | −0.14 | (−0.27, −0.01) | 0.030 |
Outcome . | Provider characteristics . | Beta . | 95% CI . | p value . |
---|---|---|---|---|
Proportion of antibiotic prescribing at baseline | Practice type (group)a | −0.14 | (−0.23, −0.05) | 0.002 |
Changes in number of antibiotics per episode | Age (≤39)b | −0.13 | (−0.24, −0.02) | 0.024 |
Age (40–49)b | −0.16 | (−0.26, −0.06) | 0.002 | |
Changes in number of antibiotics per episode | Practice location (urban)c | −0.14 | (−0.27, −0.01) | 0.030 |
areference: solo practice.
breference: age 50 and older.
creference: rural area.
The average percentage of viral illness episodes treated with antibiotics was 80.3% at baseline; this decreased by 7.6 percentage points after the policy. The average number of antibiotics was 1.35 at baseline, which decreased by 0.17 after the policy.
In multivariable regression analysis, physicians in group practices were less likely to prescribe antibiotics for viral illness than those in solo practices at baseline (−14.3 percentage points [−23.4 percentage points, −5.2 percentage points], p = 0.002). No other clinic or individual predictors were associated with tendency to prescribe antibiotics for viral illness before the policy, or with changes in the proportion of antibiotic prescribing following the implementation of the policy.
After the policy, younger physicians were more likely to decrease antibiotic polypharmacy for viral illness. Physicians under age 40 reduced prescribing per episode by 0.13 antibiotics more (0.02, 0.24; p = 0.024) than those aged 50 and older, and those aged 40 to 49 reduced prescribing per episode by 0.16 antibiotics more (0.06, 0.26; p = 0.002) than those aged 50 and older. In addition, urban location was associated with a greater post-policy decline in antibiotic polypharmacy for viral illness (−0.14 [−0.27, −0.01], p = 0.030).
Discussion
Policy effects on reducing drug prescribing
This study investigated the impact on physician prescribing of regulatory restrictions on the dispensing of medications by physicians. Unlike previous cross-sectional studies (Morton-Jones and Pringle 1993; Trap and Hansen 2002a,b; Trap et al. 2002), we examined prescribing in the same group of physicians before and after the introduction of a national policy that prohibited them from dispensing medicines. Removing the economic incentive of dispensing was found to reduce prescribing of both antibiotic and non-antibiotic medications for patients diagnosed with either viral or bacterial illness. Our results confirm the findings of previous studies that dispensing physicians prescribe more antibiotics and more drugs, and have higher drug expenditures, than non-dispensing physicians (Morton-Jones and Pringle 1993; Baines et al. 1996; Chou et al. 2003).
Office-based physicians in Korea are reimbursed on a fee-for-service basis in the national health insurance system. Before the dispensing restriction policy, all office-based physicians dispensed drugs in their clinics and were reimbursed for each drug dispensed with prices set by the government insurance system. The reimbursed prices were determined differently for each drug based on reports from pharmaceutical companies and post survey of purchase prices of sample drugs. However, the reimbursed prices tended to be much higher than the actual purchasing prices of hospitals and clinics, which enabled them to gain an additional profit from drugs dispensed. The more drugs physicians dispensed, the more profit they earned. These factors contributed to excessive prescribing in Korea. Antibiotics were commonly prescribed for non-bacterial infectious diseases (Park and Moon 1998) and gastrointestinal drugs were frequently prescribed for patients with no gastrointestinal symptoms (Byeon 1997).
The policy prohibiting physicians from dispensing was implemented nationwide by law. It no longer allowed medication dispensing by physicians, but compensated them by raising service fees to offset some of their resistance to the policy. Since physicians had no economic incentives to prescribe drugs, prescribing of both antibiotics and non-antibiotic drugs decreased after the new policy.
Policy effects on the quality of drug prescribing
We also investigated if the reduction in prescribing associated with the policy improved the appropriateness of prescribing. To do this, we examined antibiotic prescribing in viral illness cases for which antibiotics are inappropriate versus bacterial illness cases for which they may be justified. Antibiotic prescribing decreased after the policy in both groups, but the decreases were greater among patients with viral illness. In contrast, prescribing of gastrointestinal drugs, which are not related to our study illnesses, declined equally in both groups of patients in response to the dispensing restriction. Since we evaluated the same patient records in the analyses of changes in prescribing for both antibiotics and gastrointestinal medications, the selective reduction in inappropriate antibiotic prescribing when physicians were no longer allowed to dispense indicates that the physicians were aware that viral diagnoses do not require antibiotics, and the removal of the dispensing incentive encouraged them to behave more in accordance with recommended practice. Our study is the first to demonstrate that quality of physician prescribing improves when physicians are no longer allowed to dispense. The policy also decreased the number of different antibiotic and non-antibiotic drugs prescribed per patient. Although we did not analyze which classes of drugs declined the most after the policy, these findings suggest that duplicate use of antibiotics and unnecessary polypharmacy both decreased.
Despite improvements, however, inappropriate antibiotic prescribing and other drug prescribing remained high after the policy. Although prohibiting physicians from dispensing removes the possible motivation to prescribe drugs for economic reasons, the policy alone is not enough to change prescribing habits that might have been formed by other factors. Nearly half of primary physicians in Korea believe that antibiotics help treat the common cold in children (Cho et al. 2004). The fee-for-service-based reimbursement system and lack of organized programmes to reinforce appropriate prescribing also contribute to continued high rates of prescribing. Further targeted interventions will be needed to substantially reduce inappropriate prescribing and to improve physician prescribing behaviours.
Provider factors associated with a decrease in inappropriate antibiotic prescribing
We found that younger physicians were more likely to reduce antibiotic polypharmacy than older physicians when no longer permitted to dispense. Younger physicians have been reported to prescribe more appropriately than older physicians (Mainous et al. 1998; Lam and Lam 2001). Removal of the dispensing incentive may allow more clinically appropriate practices among younger physicians to emerge.
Urban location was associated with a somewhat greater decrease in antibiotic polypharmacy in our study. Physicians practicing in urban areas have better access to new information and might be more susceptible to changes in peer norms or government policies than their counterparts in rural areas. However, in another study (Trap et al. 2002), location of practice was not related to polypharmacy of dispensing or non-dispensing doctors.
Unfortunately, our data sources did not include other characteristics, such as physician education, which have been demonstrated to influence physician prescribing in other studies (Hemminki 1975).
Changes in diagnostic coding
In classifying diagnoses, we relied on physicians' judgments since we lacked supporting data such as laboratory tests. We were concerned that physicians might shift diagnosis codes after the policy to justify prescribing antibiotics, rather than changing behaviour (Finkelstein et al. 2000). Therefore, we tested whether the propensity to diagnose a patient as having bacterial illness shifted in response to the policy. First, we examined if the proportion of episodes diagnosed as bacterial or possibly bacterial among all cases of infections increased after the policy in each clinic to test the possibility of primary diagnosis shift. The ‘possibly bacterial’ diagnosis group was defined as including: bronchitis (J20, J209), acute pharyngitis (J029), and non-suppurative otitis media (H65, H651 to H653, H659, H680, H681, H690, H698, H699). Secondly, we examined whether physicians were more likely after the policy to add a secondary diagnosis of bacterial or possibly bacterial illness for patients with a primary diagnosis of viral illness.
The proportion of bacterial or possibly bacterial primary diagnoses among all episodes of infections in the study clinics did not increase (from 52.2 to 51.4%) after the policy. There was also no change in the proportion of episodes having bacterial or possibly bacterial secondary diagnoses in all viral illness episodes (from 33.7 to 33.6%).
The fact that there was no evidence of diagnosis shift strengthens the results of this study. However, the results are still limited in that we could not assess the appropriateness of the coded diagnoses because the administrative database did not provide detailed information about clinical condition. Our evaluation assumes that the relative proportion of actual viral versus bacterial illness did not change after the policy.
Limitations
This study investigated selectivity in the reduction of antibiotic prescribing for viral illness versus bacterial illness for selected illness episodes in the pre- and post-policy periods. Therefore, our analysis could not control for general pre-intervention trends, or differences in the trends in the two disease groups (Soumerai et al. 2000). However, data on previous drug prescribing supports our findings. Antibiotic prescribing for upper respiratory tract infection in Korea hardly changed between 1994 and 2000, from 85.6 to 88.7% in adults and from 90.6 to 89.0% in children (Park and Moon 1998; Jang 2001). The average number of different drugs per patient with upper respiratory tract infection was 5.1 in 1994 (Park and Moon 1998), the same as that observed for viral illness episodes before the policy studied here. Gastrointestinal prescribing, which dropped after the policy, had increased between 1997 and 2000 (HIRA 2000). These data support the assertion that the declines in use observed in this study were primarily due to the policy intervention.
The outcomes observed in this study were confined to short-term effects. Previous studies have not confirmed that changing financial incentives has a long-term impact on prescribing (Stewart-Brown et al. 1995; Wilson and Walley 1995). It is difficult to predict if the observed short-term effects on prescribing will be sustained for a longer period. Further studies using longitudinal data are needed to explore the long-term effects of this policy. We assessed just a few indicators when looking at the changes in the quality of prescribing following the new policy. Future studies should also analyze other aspects of prescribing and patient outcomes to determine other policy impacts on quality of care and how positive changes in quality could be enhanced through accompanying interventions.
Conclusions
This study demonstrated that prohibiting physicians from dispensing drugs can lead to improved quality of drug prescribing in addition to decreasing overall medication use. The new policy abruptly prohibited Korean physicians from dispensing for the first time. Our study found that such a policy can also selectively reduce inappropriate antibiotic prescribing. Many countries that allow physicians to dispense drugs continue to have problems related to inappropriate prescribing. For those countries, Korea's example can provide important evidence that prohibiting medication dispensing by physicians can contribute significantly to promoting the quality of drug use.
Biographies
Sylvia Park, Ph.D., MPH, is a senior researcher in the Korea Health Industry Development Institute, where she conducts research on pharmaceutical policy and the pharmaceutical industry. She was formerly a pharmaceutical policy research fellow in the Department of Ambulatory Care and Prevention (DACP) at Harvard Medical School and Harvard Pilgrim Health Care. Her primary research interests include pharmaceutical policies and their impacts on drug utilization, quality use of medicines, access to essential medicines and the pharmaceutical industry.
Professor Stephen B Soumerai, Sc.D., directs the Drug Policy Research Program in DACP at Harvard Medical School and Harvard Pilgrim Health Care. He co-chairs the Statistics and Evaluative Sciences concentration of the Harvard University-wide Ph.D. programme in Health Policy, and has been a member of several federal scientific review committees. He is known for his research on the effectiveness of educational, administrative and regulatory interventions intended to change clinicians' drug prescribing and other health care decisions, and for studies on the effects of drug coverage and cost-containment policies on access to effective medications, quality and costs of care, and clinical outcomes among vulnerable populations, such as chronically ill elderly.
Alyce S Adams, Ph.D., MPP, is an assistant professor in DACP at Harvard Medical School and Harvard Pilgrim Health Care. Her research focuses on pharmaceutical access issues for chronically ill patients at high risk for disability and other adverse outcomes. Her most recent publications have examined the impact of coverage and race on under-use of clinically essential medications among patients with diabetes and hypertension. Dr Adams is also Co-Director of the Fellowship in Pharmaceutical Policy Research at Harvard Medical School. She received her Ph.D. in health policy in 1999 from Harvard University.
Jonathan A Finkelstein, MD, MPH, an associate professor at DACP and in paediatrics, is a paediatrician and health service researcher at Harvard Medical School and Harvard Pilgrim Health Care. He has utilized managed care systems to test care improvement strategies for childhood asthma, and to study asthma care for low-income children. He currently leads a randomized trial of a community-based approach to promoting appropriate antibiotic use. His research interests include promoting high quality primary care practice for children in the areas of asthma care and antibiotic use.
Sunmee Jang, Ph.D., MPH, is a senior researcher at the Health Insurance Review Agency, which is a Korean national public agency to review the country's National Health Insurance. She has conducted research on evaluation of the drug policy and on drug utilization. Recently her main research topic is the introduction of a new patient classification and payment system in long-term care.
Dennis Ross-Degnan, Sc.D., is an associate professor in DACP at Harvard Medical School and Harvard Pilgrim Health Care. He is also co-Director of the World Health Organization Collaborating Center on Pharmaceutical Policy based jointly at DACP and the Boston University Center for International Health. His professional career has focused on improving health in developing countries. In 1990, he co-founded the International Network for Rational Use of Drugs (INRUD), a global network of academics, health managers and policymakers involved in developing and testing interdisciplinary interventions to improve use of medicines. His research focuses on the effects of pharmaceutical policies, factors underlying appropriate use of medicines, and interventions to improve quality of care.
We thank Ken Kleinman, Sc.D., for advice on statistical analysis.
Dr Sylvia Park was a fellow in Pharmaceutical Policy at Harvard Medical School when this work was completed. Drs. Stephen B Soumerai, Jonathan A Finkelstein and Dennis Ross-Degnan are investigators in the HMO Research Network Center for Education and Research in Therapeutics, which is supported by a grant from the US Agency for Healthcare Research and Quality (Grant No. U18HS1039–01). This work was also supported by the Harvard Pilgrim Health Care Foundation.
References
Arason VA, Kristinsson KG, Sigurdsson JA et al.
Arnold KE, Leggiadro RJ, Breiman RF et al.
Baines DL, Tolley KH, Whynes DK.
Brown SS, Surender R, Bradlow J, Coulter A, Doll H.
Byeon JJ.
Chen DK, McGeer A, de Azavedo JC, Low DE.
Cho HJ, Hong SJ, Park S.
Chou YJ, Yip WC, Lee CH et al.
Doern GV, Heilmann KP, Huynh HK et al.
Finkelstein JA, Metlay JP, Davis RL et al.
Finkelstein JA, Stille C, Nordin J et al.
Health Insurance Review Agency (HIRA).
Hemminki E.
Jang SM.
Lam TP, Lam KF.
Lavizzo-Mourey RJ, Eisenberg JM.
Lee EK, Bae JM, Park KH et al.
Lee NY, Song JH, Kim S et al.
Mainous AG 3rd, Hueston WJ, Love MM.
Morton-Jones TJ, Pringle MAL.
Nizami SQ, Khan AI, Bhutta ZA.
Park S, Moon OR.
Robbins AS, Chao SY, Fonseca VP.
Soumerai SB, Majumdar S, Lipton HL.
Stewart-Brown S, Surender R, Bradlow J, Coulter A, Doll H.
Trap B, Hansen EH.
Trap B, Hansen EH.
Trap B, Hansen EH, Hogerzeil HV.
Wang EE, Kellner JD, Arnold S.
Author notes
1Department of Ambulatory Care and Prevention, Harvard Medical School and Harvard Pilgrim Health Care, Boston, MA, USA, 2Korea Health Industry Development Institute, Seoul, Republic of Korea and 3Health Insurance Review Agency, Seoul, Republic of Korea