In the mid to late '80s, a tool called remote data entry was available which replaced double key data entry and paper case report forms (CRFs) at the clinical trial study site. When remote data entry (RDE) was used, the drug sponsor would provide a portable computer to the investigational site. The coordinator would collect study-related patient data, and then enter the data directly into the computer via the specially designed user interface of data entry screens. The electronic data would then be monitored. After data cleanup, a floppy disk with the site's data would be sent to the sponsor via an overnight courier service. These tasks would occur periodically during the course of the clinical trial.

More recently, Internet-based clinical trials (a prime aspect of electronic data capture called IBCTs for simplicity) have been introduced. However, there are major challenges for companies who choose to implement IBCTs. Some clinical research professionals question the advantages of moving away from "tried-and-true" tools such as paper CRFs. These individuals point out that such tools have worked well over time, and "if it ain't broke, don't fix it." In addition, clinical research study sites may be hesitant to embrace IBCTs as they may have to shift resources and retrain staff.

The reality is that paper-based systems do not work very well and are not efficient. Paper-based systems, by definition, generate enormous amounts of paper, involve redundant and inefficient processes, and introduce the potential for major delays from the time of data capture to the time the data can be reviewed and analyzed.

Figure 1. Sample data entry screen.

A recently published summary revealed practical considerations and steps in setting up IBCT.⁴ Another paper⁵ demonstrated the areas of cost saving in IBCTs and how the query rate is markedly reduced compared with a paper-based trial. In terms of data quality, it was also shown⁶ that data capture in a IBCT had a very low error rate, and that the errors observed were related to human activities and not related to the underlying technology. Banik⁷ showed that queries in electronic data capture-based studies can be reduced by at least 80%, compared to paper-based queries. Banik also showed that compared to paper-based studies, IBCTs can reduce clinical trial duration by at least 30%, time to database lock by 43%, and queries by 86%.

The changing role of data managementData entry. In a paper CRF trial, the study site coordinator completes a paper CRF from data documented in the patient record. When the monitor arrives at the study site, the contents of the CRFs are reviewed for accuracy against the patient record. Once the data transcription is verified, the monitor "pulls" the CRF pages and either overnights them or physically brings them to the office upon return from the monitoring trip. These paper CRFs have been known to get lost and/or damaged every once in a while.

Once in-house, the pages are sent to the data entry group, which logs them in and enters the data into each of two parallel databases. Each "double entry" is then compared, usually using a SAS-generated data-compare program. If the two entries correspond to each other, the data are accepted. However, if the two entries do not correspond, the correct entry is determined by referring to the original CRF and the incorrect entry is changed. If the data entry clerk cannot read a field, or the data seem illogical or obviously incorrect, a notification is sent to the clinical research associate (CRA) and a query is sent to the site to clarify or correct the data entry field in question. Finally, data listings are generated in SAS, and quality control (QC) procedures are put in place to verify that the data listings are consistent with the original paper CRF. At the time of the final QC, all paper queries must be filed with each CRF to ensure that the final database reflects both the data captured on the CRFs and the resolved queries.

Figure 2. Sample data entry screen with identification of missing data.

Upon electronic submission of the data to the database, the system automatically checks for inconsistent, illogical or missing data, and when appropriate, sends back an error message notifying the site of the possible error. Figure 2 illustrates a "soft" edit check in which missing data are flagged; in this case "Race" was missing. The system identifies the missing data element and prompts the coordinator to explain why Race is missing.

Data manager. While data management responsibilities vary between companies, in general the data manager acts as a bridge between clinical research and biostatistics. In paper CRF clinical trials, the CRFs are usually prepared by the clinical group. The data manager may or may not see the final CRF. This gap may be obviated, at least in part, if a company has worldwide standards for CRF and database design. However, in our experience, this is not usually the case. As a result, at some point in time the data managers receive a copy of a CRF, for which they may have had no impact on design. The CRF is then annotated to describe the variable names, their definitions, any characteristics of the variable such as numeric versus text, the length of the variable, and any specific decodes such as a yes/no drop down box. After CRF annotation, the data manager may be involved in database design, and also the generation of SAS edit checks in order to look for data entry errors and discrepancies. In some circumstances, the data manager can even generate queries to resolve data inconsistencies in addition to creating the data listings. The data manager states when the database is clean and is directly involved in the tasks surrounding database "lock."

In IBCTs, the annotated CRF, database design, and the vast majority of edit and logic checks (some SAS edit checks are still needed) are finalized prior to enrolling the first subject. In other words, most of data management planning and implementation are completed even before data entry is initiated. Operationally, the data manager now reviews the CRF prior to its finalization and provides input to the clinical group in order to allow for CRF and database optimization. While the data manger still generates the edit checks for implementation by the programming group, this task must be done immediately after CRF design is completed. The reason is that edit check specifications must be given to the programmers for incorporation into the data entry forms prior to initiation of the clinical trial. Each form can now be "initially reviewed" by the CRA at the time of source document verification. The data manager can now look at the data in real-time, lock the form after final review, have rapid access to SAS datasets, and begin the generation of data listings and identification of potential problems early during the course of a study.

Query management. Query management is one of the main tasks in data "cleanup." Queries are usually generated by CRAs and/or Data Management. Medical monitors may also request clarifications from a CRA during medical review of the data. Currently, query management is a paper-intensive and labor-intensive process. Queries are usually generated manually by completing a two-part no-carbon-required paper form. The form is sent to the clinical study site for resolution by the coordinator and approval by the investigator, and finally returned to the sponsor. The query is then filed with the appropriate paper CRF.