The Factors Of Clinical Data Quality And The Selection Of Data Collection Method

Objective:The purpose of this study is to explore the significant factors that influence error rate and handling time of clinical data, compare differences of error rate and handling time between electronic data capture (EDC, E) and paper data capture(PDC, P) under different phases, regions, data collection modules(DCM) and clinical trial centers. The study is also to compare the differences of data distribution between the two data collection methods in order to provide experimental suggestions on choosing clinical data acquisition method, and summarizing the best strategy on data monitoring and managing.Method:①The clinical data is collected from a large pharmaceutical company’s database, including164projects from phase Ⅰ to phase Ⅳ in metabolic disease field. The basic information include duration, region, center number, data collection method, phase, completion status, DCM, therapeutic area, indication, drug name, and clinical trial ID.②The two evaluation indicators are defined as error rate and handling time. The data is categorized, processed, and computed according to data collection method, phase, region, DCM, and center number.③The general linear model is used to select significant factors that influence error rate and handling time which has been transferred by BOX-COX transformation. Then non-parametric test is used to analyze the specific situation of these two indexes, combined with data distribution and statistics to give comprehensive conclusion.Result:①The collection method, phase, DCM and region have significant effects on error rate; collection method, phase, region and center number are four significant factors on handling time. As for data collected by EDC, three significant factors that affect error rate are phase, DCM and center number; four significant factors that affect time are phase, DCM, region and center number. For data collected by PDC, phase, DCM and center number significantly affect error rate; clinical period, region and center number significantly affect handling time.②Overall, the error rate of EDC(mean:8.37%, median:3.45%) is larger than PDC(6.27%,2.87%), the query handling time of EDC(14.44days,9.52days) is much shorter than PDC(53.25days,32.22days). The error rate distribution between EDC and PDC in ten DCMs are disorderly. Compared with PDC, EDC has higher error rate and more outliers in phase Ⅰ, higher error rate and less outliers in phase Ⅱ, no significant difference of error rate distribution in phase Ⅲ and phase IV, but slightly lower value of statistic including mean and median in phase Ⅲ and higher value in phase Ⅳ. Error rate of data collected in Asia and Latin America by EDC is higher than PDC but less outlier, error rate distribution of Europe and North America is not significantly different, for value EDC of Europe is higher and PDC of North America is higher. The difference of error rate distribution between EDC and PDC is not significant in both single center and multi-center, and PDC value of single center is higher and EDC in multi-center has higher value. Except phase Ⅰ, the handling time distribution of all DCMs, regions, phases and centers present significant difference between EDC and PDC, and PDC has larger value and more outliers.③For PDC data:the error rates of COMT (17.13%,14.60%), MEDH (11.89%,10.29%), ADVE (7.41%,5.04%), DRGR (7.17%,3.25%) are larger than other modules and intensively distribute in higher area. There is an increasing trend of error rate from phase Ⅰ to phase Ⅳ; the query handling time in phase Ⅰ (13.63days,6.91days) is the shortest, phase Ⅲ (76.81days,45.98days) is the longest. The error rates of Europe (6.93%,3.32%) and Latin America (6.74%,3.76%) are higher, Asia (3.43%,0.89%) is the lowest; Latin America (99.38days,58.25days) uses the longest time and Asia (22.00days,19.30days) the shortest. The time cycle of multi-center (77.20days,47.96days) is longer than single center (33.37days,21.78days).④For EDC data:the modules of COMT (23.03%,20.42%), MEDH (25.04%,20.73%), ADVE (11.85%,8.74%), DRGR (7.81%,2.14%), DEMG (7.68%,5.79%) have higher error rate and higher regional distribution; the handling time of LAB (26.01days,13.36days) is longer than other modules. Phase Ⅱ has higher error rate (9.82%,5.80%), phase Ⅰ (7.86%,2.65%) is lower but more outlier. Latin America (25.41days,21.54days) has the longest handling time whereas Asia (7.98days,6.05days) is the shortest. The error rates of multi-center (9.80%,4.92%) is larger than single center (7.81%,2.71%), and time (21.88days,16.77days) is also longer than single (11.47days,6.78days).Conclusion:①Data collection method is a significant factor that causes different error rate and query data handling time. This conclusion is consistent with the thesis expectation. Compared with PDC, EDC is more vulnerable to more factors. From data collection to management, the effect of collection method, phase, region and DCM on error rate should be considered in a comprehensive way; the effect of collection method, phase, region, and center number on handling time should also be considered.②In general, PDC takes the advantages of improving data quality and stability; EDC takes the advantages of shortening trial period and controlling outliers. DCM is not valuable for choosing method. For phase Ⅰ and Ⅱ clinical trial, EDC is better to shorten time if budget is sufficient but data quality control should be strengthened such as optimizing the electronic system function; otherwise PDC is better in improving data quality. For phase Ⅲ, EDC is much better than PDC in controlling data quality, reducing outlier and shortening time; and phase IV, method should be chosen according to the practical situation such as duration, region, patients’ number, trial design and so on. EDC can be used for better controlling outliers and shortening trial cycle in Asia and Latin America, but a little weak at controlling quality and stability. The project budget and medical condition should also be considered when making choice. Due to many countries involved in Europe, method choosing should take others factors into account. For North America, EDC is more suitable. Single center trial tends to select EDC whereas multi-center trial is PDC.③If PDC is chosen, data monitoring of COMT, MEDH, ADVE, DRGR modules, phase Ⅱ and Ⅲ trials, Europe and Latin American trials should be enhanced; other measures to take into account include improving the protocol compliance and training, absorbing data managing experiences from Japan, raising responsibility of medical personnel, avoiding missing data to the great extent, simplifying query process in later phase, Latin America and multi-center trials, and increasing the response and feedback speed to query.④If EDC is chosen, data monitoring of COMT, MEDH, ADVE, DRGR, DEMG modules, phase Ⅱ trial and multi-center trial should be enhanced; other measures to take into account include simplifying query process in later phase, Latin America and multi-center trial, improving the consistency of implementing standard and speed of responding to query to decrease longer query dealing time in LAB module; training, system function setting, and safety maintaining should be well prepared to reduce large number of outliers in phase Ⅰ.