Data Sources
This study used data from the Centralized Drug Procurement Survey in Shenzhen 2019 (CDPS-SZ 2019) [17]. In China, the CDPS-SZ 2019 was organized and conducted by the Global Health Institute of Wuhan University between December 2019 and January 2020. The survey aimed to evaluate the effect of drug-related policies in Shenzhen, China, and collected monthly drug purchase order data between 2018 and 2019. In the CDPS-SZ 2019 database, each purchase order record included purchase date, generic name, dosage form, specification, pharmaceutical manufacturer, price per unit, purchase volume, purchase expenditures, etc. A general database containing 963,127 monthly aggregated purchase order records was established, involving 1079 drug varieties (by generic name), 346 medical institutions, 857 pharmaceutical manufacturers. The total purchase expenditures reached RMB 20.87 billion.
The purpose of this study is to examine the impact of NCDP policy on prices of policy-related drugs. Thus, we included samples with the following criteria: (a) the medication covered by “4+7” CP and their alternative drugs (Table A1). The CP drugs were sorted into winning and non-winning products based on the bidding results of "4+7" policy; (b) the time period between January 2018 and December 2019; and (c) the medical institutions purchasing drugs were from Shenzhen, China. Finally, 47,163 purchase order records of 82 drug varieties (by generic name) were included in the analysis. Figure 1 presents the flow chart of sample screening.
Outcome Variables
Drug Price Index was used as outcome variable in this study, which is a common indicator reflecting the trend of drug price change over different periods [20, 21]. This study applied three commonly used DPI: Laspeyres Price Index (LP), Paasche Price Index (PP), and Fisher Price Index (FP).
LP is calculated as the ratio of price in reporting period and the price in baseline period, weighted by the quantity in baseline period. This method assumes that the consumption structure of drugs remains unchanged in different periods, and is applicable to reflect the pure price change of drugs. LP is calculated as follows:
PP is calculated as the ratio of price in reporting period and the price in base-period, weighted by the quantity in reporting period. The index reflects the increase or decrease in drug costs due to the price change when the quantity and consumption structure has changed. PP is calculated as follows:
FP is calculated as the geometric mean of LP and PP, which was weighted by the quantity in both baseline period and reporting period. Thus, FP can equalize and average the biases of LP and PP. Theoretical researches showed that FP is an optimal form of price index, and is called “the ideal index” [22]. FP is calculated as follows:
In the above formula (1)-(3), P means the price, P0 and P1 refer to price per Defined Daily Doses (DDDs) of each product in baseline- and reporting period. Q means purchasing quantity, Q0 and Q1 refer to DDDs of each product in baseline- and reporting period. DDDs is the ratio of the quantity of drug utilization and Defined Daily Dose (DDD) [23]. If the drug price index > 1, it indicates the increase of drug price in the reporting period when compared with the base-period; If the drug price index =1, it means that drug prices remain unchanged over the two periods; If the drug price index < 1, it means that drug prices in the reporting period decrease compared with the base period.
In this study, January to June 2018 was assigned as the baseline period, and July 2018 to December 2019 was assigned as the reporting periods to calculate drug price indexes of each month (18 months).
Statistical Analysis
Descriptive statistics were used. We first described the change of DPI of included drugs before (July 2018 to February 2019) and after (March to December 2019) the implementation of "4+7" policy.
Interrupted time-series analysis was applied to assess the change of Fisher Price Index after the implementation of "4+7" policy. ITS is a commonly used approach for evaluating changes in longitudinal series following a quasi-experimental intervention occurring at a fixed point in time. The interrupted time series was constructed with DPI data in Shenzhen from July 2018 to December 2019. The time unit was set to 1 month and the intervention time point was set to March 2019, making 18 time points available for analysis, including 8 points as pre-"4+7" policy period and 10 points as post-"4+7" policy period. To estimate the effect of the intervention on the outcome variables, the following segmented linear regression model was developed [24-26]:
Where, Yt is the independent outcome variable (i.e. Fisher Price Index) in month t. time is a continuous variable indicating time in months at time t from the start of the observation period. intervention is an indicator for time t in the pre-"4+7" policy period (intervention = 0) and post-"4+7" policy period (intervention = 1). time after intervention is a continuous variable indicating months in the post-"4+7" policy period (time in the pre-"4+7" policy period is coded 0).
In this model, β0 estimates the baseline level of the independent variable at the beginning of the observation period. β1 estimates the linear trend during the pre-intervention period. β2 estimates the level change in the outcome variable immediately following the intervention. β3 estimates the trend change in the outcome variable in the post-intervention period compared with the pre-intervention period. εt is an estimate of the random error at time t. Durbin-Watson test was performed to test the presence of first-order auto-correlation (a value around 2 indicates no sign of auto-correlation). If autocorrelation is detected, the Prais-Winsten method was applied to estimate the regression. Stata version 16.0 was used to perform the ITS analysis. A p-value <0.05 was considered statistically significant.