A study published today in BMJ Open highlights one often overlooked benefit of data sharing, the discovery of honest errors which allow correction of the published scientific record.
Data is the foundation of all medical research and data sharing, making the raw data opening available, is a cornerstone of the open science movement. Data sharing has the potential not only to improve transparency and reproducibility of research, but to drive discovery and clinical practice and reduce research waste. These anticipated benefits have led to many funding agencies and journals implementing policies that support (or in some cases mandate) the sharing of raw data in a timely manner after study completion.
To explore the potential benefits of sharing data from clinical trials, The New England Journal of Medicine and the National Institutes of Health (NIH) launched a data analysis contest in November 2016. The SPRINT Data Analysis Challenge gave researchers around the world access to the anonymized data set underlying the SPRINT trial, an NIH funded randomized-controlled trial which compared intensive management of systolic blood pressure with standard management.
After accessing the SPRINT data, Dr Krumholz and colleagues sort to replicate the published findings of the SPRINT trial. The participants of the SPRINT trial where included in the trial based on having an increased risk of cardiovascular events, and one of the measures used to assess this risk was the Framingham Risk Score (FRS), a commonly used model to estimate the chances of an individual developing cardiovascular disease over the proceeding 10 years. When calculating the FRS for the trial participants from the raw trial data the researchers discovered a discrepancy; the FRS values obtained when inputting the patients’ individual variables into the model were not the same as the FRS values provided in the raw data, or those in the published results. The researchers calculated that 76% of the SPRINT trial participants had an FRS of ≥15%, as opposed to 61% reported in the published trial results. As a result, it appeared that the 10-year cardiovascular risk of the SPRINT participants had been underestimated and many of the participants could be considered as high-risk.
As the study authors point out, their experience of data sharing highlights a key benefit of the practice, that outside researchers can reproduce existing analyses, and in that process, discover any errors in the published data. They are also keen to point out that their analyses do not change the results of the SPRINT trial. The original SPRINT publication has now been corrected.