Can an online clinical data management service help in improving data collection and data quality in a developing country setting?

Data collection concerning medical needs is required to assess the effectiveness of interventions and current health care practices [1]. Furthermore, data collection by Electronic Medical Record (EMR) systems has been proven to be helpful in data collection for scientific research and can be helpful in improving healthcare. These EMR systems allow for the early identification of missing data and the patients possibly loss-to-follow-up, which is essential for the conduct of proper scientific research [2‐6].

A Clinical Trial Data Management service (CTDMS) has been introduced for running a multicenter clinical trial in Indonesia and in the Netherlands. The same system has also been introduced for monitoring treatment results of Nasopharyngeal Carcinoma (NPC) in Indonesia.

In most countries NPC is an orphan disease, but overall has a worldwide incidence of 80.0000 new cases per year, being endemic in Northern Africa, Southern China and Hong Kong, and the South-East Asian peninsula, including Malaysia, Vietnam, Thailand, Singapore and Indonesia.

In Indonesia NPC is the most frequent cancer in the head and neck area and ranks as the 4^th most common tumour found in males. The incidence is estimated 6 per 100.000, leading to 12.000 new cases per year [7, 8]. Little is known about treatment results of NPC in Indonesia.

The CTDMS system was selected because of the web-based nature which makes the data approachable for all participating parties. This online accessible data system has made it easier for the principal investigator to check the data for inconsistencies. The senior physician can easily see if treatment is according protocol.

This study assesses whether the introduction of CTDMS composed of online Case Report Forms (eCRF) can result in improved patient outcomes. The assessment focuses on data quality and the identification of possible bottle necks within the patient care process.

This study investigates if a web based CTDMS can be helpful in proper data collection by analysing errors in data items. Bottle necks in patient care are analysed by comparison of treatment plan and actual treatment.

The CTDMS is constructed for the NPC Clinical Trial: Early detection of primary and recurrent NPC using (anti-)EBV based tumour markers and evaluation of primary treatment for NPC (funding KWF NKI-2008-4233). A technical description of the CTDMS is provided in Appendix 1. The database is comprised of 10 online eCRF's. In order to prevent errors from being entered, data validation rules were implemented into the eCRF's prior to commencement of the NPC Clinical trial. These data validation rules assess whether certain pre-specified conditions are valid and can therefore pin-point omissions or erroneous data. Online warning messages notify the data-manager (entering data) when errors are detected. Commonly used checks are, for instance, range checks that verify whether values are within the boundaries dictated by the study protocol, and mandatory field checks (i.e. 'This field cannot be blank').

Of the 10 eCRF's, 9 were required to be completed multiple times per patient during the study and only 1 was to be completed and submitted once per patient. Each of these submissions is a unique realization of the form. For example, for one patient a laboratory form is completed during baseline measurements, just before the start of the treatment. Once this form is submitted through the CTDMS, there is one realization of the laboratory form stored in the database for this patient. After the patient received treatment, a laboratory form is completed and submitted again. The data base then contains two realizations of the laboratory form, for that patient. Each realization may be submitted multiple times if it contained errors. We note that it is impossible to claim that an entered form for which no warning messages were displayed is clean, as new errors may be found later.

The data-manager completing an eCRF has the option to ignore (override) a warning message, however in such cases, he/she is required to provide an explanation which is recorded in an Audit Trail entry field (error log). Warning messages and error logs are also created when an incorrect value or data-type is entered, an omission is detected, or when a previously entered value is changed. Changed values are considered to be (previously undetected) errors that have now been rectified (except when the changed value also triggers a check to fire, in which case the data is considered unclean).

The eCRF's contain differing quantities of data. Each field to be entered is considered a data item, which were designated either as primary or secondary. Primary data items are data that were considered essential for the assessment of the NPC Clinical trial primary endpoint, and so for assessment of treatment protocol. Secondary data items are data required to assess the clinical trial's secondary endpoints.

As acceptable levels of data quality, an 1% error rate for primary and an 2.5% error rate for secondary data points were adopted [9]. We present the change in error rate over the course of the trial, the number of errors per submission, and the change in data quality per form per submission.

Between November 2009 and March 2010 a total of 4860 data items pertaining to 51 patients were entered. This is the first five months of an estimated 3 year long accrual period. In total 433 eCRF's were submitted, of which 329 were unique realizations. Each CRF has been submitted between 1 to 4 times. Table 1 presents an overview of the submitted eCRF's and data items.

Of the 433 submitted eCRF realizations, 287 were submitted for the first time without primary data errors (Table 2), while 253 forms (realizations) were submitted for the first time without secondary data errors (Table 3). No form had more than two errors in the primary data. One form contained 10 secondary data errors when it was submitted for the first time. This was baseline patient registration data for which the wrong patient was entered. In general subsequent submissions contained fewer errors (Figure 1).

For example, the "Pathology, Staging & Given treatment" eCRF contained a total of 89 unique data items, with the number of data items per eCRF ranging from 2 to 18 (Table 1). Of these 89 items, 40 were classified as primary data, with the remaining 49 being classified as secondary. The error rate at first submission was 3.3% for primary data and was 8.4% for secondary data.

To assess the change in data quality over time, the proportion of unsolved errors in primary and secondary data were plotted against time (in months). Figure 2(A) presents the cumulative number of unique data items submitted and the number of unresolved errors over the first five months of the study. Figure 2(B) presents the change in the percentage of unresolved errors of primary and secondary data items. Although the absolute number of unsolved errors is increasing with time (due to the accrual of patients), the fraction of erroneous data is declining. Five months after the start of study the error rate for the primary data items was 1.6% and for the secondary data items the error rate was 2.7%. Although not quite at the levels appropriate for final analysis, the standard of data quality is high, very early into the study.

For this study we found an error rate of 1.6% for primary data items, while in earlier studies in the same setting data could not be analysed because of the massive data loss and poor data quality. With this real time data monitoring and inbuilt checks we have realized acceptable levels of data quality. The CTDMS prevents us from missing data or ending up with poor quality data at the end of the study, which often at that point cannot be resolved anymore.

The presented analysis shows that after five months since the introduction of the CTDMS the error rates for both Primary and Secondary data items reflect acceptable levels of data quality. Furthermore these error rates were decreasing over time. The drop in errors per form with each form submission indicates that, while being prompted by the CTDMS, the data manager and responsible doctors are actively solving the errors. Online warning messages notify the data manager (entering data) when errors are detected, allowing them to immediately correct the data, rather than the usual delay associated with paper based CRFs.

Clearly, the CTDMS encourages local data managers to verify the entered data and, if necessary, ask the doctor whether the information is correct. It is also likely that the reason that data managers have to specify arguments before submitting the form in case the CTDMS detects erroneous data motivates them to verify whether the available data is actually correct. This may explain why our results show a significant increase of clean data and a self-learning curve of the data manager is to be expected. Moreover, the error logs provide valuable information about the bottlenecks in the treatment of the NPC patients.

In the past authors have pointed out that existing data collections in developing countries are often deficient [10, 11]. Eiseman and Fossum (2005) emphasize that existing data collections are insufficiently comprehensive, sometimes inaccurate, and often out of date by the time the data can be acted upon. All point out that without these data the required empirical knowledge to address the health problems in developing countries is insufficient. Especially on strategic planning, priority setting, monitoring and evaluation, advocacy, and general policymaking [12‐14].

These comments supported us on introducing an online medical record system which could play an important role in improving data collection and data quality. Accordingly, during analysis we have also seen that treatment procedures are often unsatisfactory. The first analysis regarding the treatment of NPC has been presented and discussed with all members of the disciplinary team. The main concern was the duration of the radiotherapy. According to the protocol the duration for administering the 66 gray radiotherapy should take to the utmost 42 days, yet analysis showed that the treatment time takes in average 66 days, which will lead to inadequate treatment [15, 16]. Future analysis has to show if intervention by CTDMS system-based education of the doctors will eventually lead to better treatment outcome. The digital nature of the CTDMS, as well as the online availability of data, gives fast and easy insight in adherence to treatment protocols. As such, the CTDMS can serve as a tool to directly train and educate medical doctors. Therefore, a potential even bigger advantage of an online medical record system is the ability to monitor the data from the teaching hospitals especially in developing countries. Via this way the teachers can communicate directly or visit the participating hospitals with a custom fit teaching program, which will make such visits more effective.

We show that an online clinical data management service can improve data quality in a developing country setting. In the future we expect to see both less loss-to-follow-up and better treatment programmes with help of this CTDMS. For better and more efficient medical care programs and studies in developing countries we believe an online data system is essential.

The digital nature of the CTDMS, as well as the online availability of that data, gives fast and easy insight in adherence to treatment protocols. As such, the CTDMS can serve as a tool to train and educate medical doctors and can improve treatment protocols. Since the introduction of this system training the doctors has become much more efficient.