Design overview

We defined our study cohort based only on in-hospital coronary catheterisations, as these procedures are not commonly performed on an outpatient basis in China. We included patients undergoing coronary catheterisation for any indication. We sampled eligible hospitalisations for 2001, 2006 and 2011 from a network of nationally representative hospitals. We chose these three time periods to reflect the influence of major changes in the Chinese healthcare system and the creation of PCI guidelines by Chinese medical societies (figure 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Chinese trends in PCI volume from 2001 to 2011. Figure demonstrates the increase in PCI volume by year and the increase in the number of hospitals performing >100 PCIs per year. Notable events pertinent to the expansion in PCI access with time are highlighted below the graph. In 2003, the Rural Cooperative Medical Care System expanded health insurance to low-income rural residents. The healthcare reform of 2007 substantially expanded health spending for the rural and urban population. We sampled eligible hospitalisations for 2001, 2006 and 2011. PCI, percutaneous coronary intervention.

To study 10-year trends in patient characteristics, treatment patterns and outcomes nationally and within regions of different socioeconomic development, we drew a representative, stratified sample of patient discharges for each year. We intentionally drew a larger sample for 2011 to obtain more precise estimates of the current performance and variation in treatment patterns and outcomes among hospitals.

The central ethics committee at the China NCCD approved the China PEACE-Retrospective CathPCI Study. All collaborating hospitals accepted the central ethics approval except for five hospitals, which obtained local approval by internal ethics committees. The study is registered at http://www.clinicaltrials.gov (NCT01624896).

The Chinese government, which provided financial support for the study, had no role in its design or conduct; in the collection, management, analysis and interpretation of the data; or in the preparation or approval of the manuscript.

Sampling design

We chose hospitals to reflect the diverse sites of care that perform coronary catheterisation and PCI. Candidate hospitals were limited to urban areas, as catheterisation capability is restricted almost exclusively to these regions. We identified urban areas using official administrative divisions, in which a region is considered urban if it is part of a downtown or suburban area within a direct-controlled municipality (Beijing, Tianjin, Shanghai and Chongqing) or 1 of 283 prefectural-level cities. In total, Mainland China is composed of 287 urban regions. As financial and medical resources are not identical throughout Mainland China, we separately identified hospitals in each of its three official economical-geographical regions, that is, Eastern, Central and Western. As Central and Western urban regions have similar per capita income and health services capacity, we combined Central and Western regions into one stratum.18 We then identified hospitals separately for two study strata: Eastern-urban and Central/Western-urban.

We identified cases for study inclusion using a stratified two-stage cluster sampling design (figure 3). In the first stage, we identified hospitals using a simple random sampling procedure in each stratum. The sampling framework consisted of the highest-level hospitals in each of the predefined urban regions with the potential capability of performing PCI (833 hospitals in 287 urban regions). Hospital level is officially defined by the Chinese government based on clinical resource capacity, for example, secondary hospitals have at least 100 inpatient beds and the capacity to provide acute medical care and preventive care services to populations of at least 100 000, while tertiary hospitals are large referral centres in provincial capitals and major cities.15 ,19 We excluded military hospitals, prison hospitals, specialised hospitals without a cardiovascular disease division, and traditional Chinese medicine hospitals. We selected representative hospitals from 2011 to reflect current practices and traced this same hospital cohort backwards to 2006 and 2001 to describe temporal trends. As hospital numbers have grown by approximately 18% over the past decade,20 ,21 the cohort should be most representative of national treatment patterns and outcomes in 2011.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

The China PEACE-Retrospective CathPCI Study flow chart and associated quality assurance strategies. The flow chart should be read from top to bottom. CRF, case report form; Q & A, questions and answers.

In the second stage, we drew cases based on the local hospital database for patients who underwent coronary catheterisation or PCI at each sampled hospital using systematic random sampling procedures. For each stratum, we determined the sample size required to achieve a 1.5% precision for describing the percentage of patients experiencing in-hospital complications, which we estimated at 5%.9 ,22–25 To achieve a precision of 1.5% with an α of 0.05 in each stratum, assuming an intraclass correlation of 0.02 and design effect of 2.2, we would need to sample 1750 records among hospitals with an average cluster size of 60. This cluster size appeared reasonable based on national administrative data26 and our previous survey of treatment for acute coronary syndromes at more than 1000 hospitals in 2010, which demonstrated that the median annual PCI volume was approximately 300 cases. Assuming a participation rate of 85% among selected hospitals, we approached 35 hospitals for participation in each stratum, for a total of 70 hospitals. We doubled cluster sizes for 2011 to improve precision in the description of hospital-level treatment patterns and outcomes. Consequently, the total expected sample volume with the above assumptions was approximately 3500 cases in 2001, 3500 cases in 2006 and 7000 cases in 2011. A more detailed description of the sampling strategy is provided in the supplemental material within the online web appendix.

Data collection

We trained staff at participating hospitals to identify all hospitalisations with coronary catheterisation or PCI procedures from their respective local hospital databases for 2001, 2006 and 2011. After we sampled cases at each hospital, we assigned each case a unique study ID. We then required local investigators to obtain the original record and transmit a scanned copy to the coordinating centre. To facilitate this process, the coordinating centre provided each study site with a high-speed scanner. To verify compliance with the case-finding strategy, research staff from the coordinating centre visited 60% of the sites to repeat the case-finding process, confirm that the list of hospitalisations undergoing coronary catheterisation or PCI was complete, and assist in acquiring the sampled cases (figure 3). We chose to visit study sites that were expected to contribute a large volume of cases and have potential difficulties in complying with the case-finding strategy. These sites provided 73% of sampled cases.

Following receipt of the scanned record, research staff at the China NCCD ensured the completeness and quality with which each medical record was copied. We required incomplete or poorly scanned records to be rescanned and retransmitted (figure 3). We instructed study sites to include all parts of the medical record, including the face sheet, admission note, daily progress notes, procedure notes, medication administration record, diagnostic procedure reports, laboratory test results, physician orders, nursing notes and discharge summary.

The China PEACE-Retrospective CathPCI Study adhered to rigorous standards for abstraction. Before initiating the chart review, each abstractor received 2 weeks of training that included an introduction to the study, coronary heart disease and its subtypes, components of the inpatient medical record including specialised sections such as catheterisation reports, and the China PEACE-Retrospective CathPCI Study data dictionary. We provided all material, including the data dictionary, in Chinese. After training, we certified individuals who were able to abstract five sample medical records with greater than 98% accuracy. Inexperienced abstractors began with exclusively typewritten rather than handwritten medical records. In addition, we randomly audited approximately 5% of the abstracted records. If the records were not abstracted with 98% accuracy, all medical records in the audited batch were considered unqualified and were re-reviewed by a different abstractor. Abstractors were required to maintain this benchmark of 98% accuracy to retain certification. We used abstractors with formal medical training to identify data elements requiring more advanced medical knowledge, such as the development of postprocedural complications including bleeding or cardiac tamponade. A physician was always present in the room with abstractors or was available online to answer questions as they arose. We assigned medical records belonging to the same hospital and year to a broad group of reviewers to avoid potential residual disparities in quality among abstractors.

Data management

We systematically perform ongoing data cleaning. Data managers regularly query data for invalid and illogical values as well as for duplicate record entries. They identify potential invalid values by searching for outliers in continuous data distributions. Records with identical study identification numbers, hospital identification numbers, medical record identification numbers and dates of discharge trigger a search for duplicate records. Once a data query is made, concerns are resolved after tracing and reviewing the relevant records.

All data have been treated as protected health information and are thus securely stored in an encrypted and password protected database at the main coordinating centre.

Data elements

We examined the English language and Chinese literature for relevant studies to create a list of candidate variables. We supplemented these elements with variables used in the CathPCI Registry of the American College of Cardiology National Cardiovascular Data Registry (NCDR). CathPCI is an outcome-based registry and quality improvement programme that focuses on patients who undergo coronary catheterisation and PCI in the USA. Incorporating variables from CathPCI will permit cross-country comparisons in resource utilisation and adherence to quality metrics. We also incorporated elements from the case report form used in the China PEACE Retrospective Study of Acute Myocardial Infarction27 (clinicaltrials.gov identifier NCT01624883) to permit comparison of hospitalisations for acute myocardial infarction between these two China PEACE studies. Major categories of data elements are described in table 1.

In table 2, we show data relevant to performance measures during the first 24 h of hospitalisation and at hospital discharge for patients with acute myocardial infarction and those undergoing PCI for stable coronary artery disease.

Where possible, we collected data that would allow us to construct the core quality measures for acute myocardial infarction used and reported by the Centers for Medicare & Medicaid Services in the USA as well as the quality measures for PCI from the NCDR.28 Two physician investigators at each participating hospital also completed a survey, modelled on the annual survey of hospitals performed by the American Hospital Association, of its major structural and organisational characteristics during the study period.29 Key variables assessed include bed size, teaching status and capability of performing coronary artery bypass graft surgery.

Statistical analyses

We will report summary statistics for patient characteristics, use of diagnostic tests, treatments received and in-hospital outcomes, including complications of care and hospitalisation costs across study sites. Weighting will reflect the reciprocal of sampling probability. For each aim, we will use standard parametric and non-parametric techniques for observational data, including t tests, χ² tests, Wilcoxon rank sum tests and generalised linear models. Because patient characteristics, treatments and outcomes may be correlated within hospitals, analyses will account for the effect of clustering. To examine and adjust for differences between comparison groups, we will use linear, logistic, Cox proportional hazard and Poisson models with a generalised estimating equation approach and hierarchical models, where appropriate. We will develop models to stratify patients according to their risk of adverse outcomes. We will assess the relationship of candidate variables to in-hospital outcomes using appropriate statistical techniques for the dependent variable. We will further refine the list of candidate variables based on their clinical relevance.

Progress to date

As of December 2012, 55 hospitals agreed to participate in the study. Fifteen hospitals did not participate because they did not provide inpatient services for coronary heart disease (5 hospitals), were incapable of performing coronary catheterisation and PCI throughout the study period (8 hospitals), or refused to participate (2 hospitals). Of the 55 participating hospitals, 29 were located in the Eastern economical-geographical stratum and 26 were located in the Central/Western economical-geographical stratum. The distribution of sites by province and region is shown in figure 4.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Geographical distribution of participating hospitals in the China PEACE-Retrospective CathPCI Study. Of the 70 sampled hospitals, 15 were unable or unwilling to participate and 55 provided cases for the study.

In parallel with national trends, the number of hospitals providing cases increased with each subsequent year of study. In 2001, 26 of the 55 participating hospitals were capable of performing PCI, of which 24 submitted medical records. Of the two hospitals that did not submit records, one did not keep records in 2001 and the other had its records destroyed by a fire. Both of these institutions most likely had a very limited capacity for PCI in 2001, as evidenced by their performance of only 16 and 28 respective cases of coronary catheterisation or PCI in 2006. A total of 44 hospitals performed PCI in 2006, all of which submitted records to the study. Fifty-four hospitals performed catheterisation and PCI in 2011. As with 2006, all hospitals that performed PCI in 2011 submitted records.

We collected all medical records for abstraction by December 2012. The examination of census databases from participating hospitals yielded 58 008 hospitalisations for coronary catheterisation and PCI (3270 in 2001, 12 875 in 2006 and 41 863 in 2011). Of these 58 008 hospitalisations, we sampled 12 477 (22%; 1444 (44%) in 2001, 3046 (24%) in 2006 and 7987 (19%) in 2011). Of these 12 477 sampled hospitalisations, we acquired medical records for 11 900 (95%); 577 (5%) medical records could not be found due to poor archiving. We began data abstraction of these medical records in March 2013 and expect abstraction to be completed in January 2014.