BACKGROUND: The WSP tool has previously been proposed as a method to detect signals for adverse drug reactions utilising time-to-event data without the need for a reference population. The aim of this study was to assess the performance of the tool on two well-known and two suspected adverse drug reactions for bisphosphonates that varied in both frequency and accuracy of reporting time. METHODS: The use of the WSP tool was investigated on data from a matched population cohort study involving data from UK primary care patients exposed to oral bisphosphonates. Four listed/suspected ADRs were selected for investigation: headache, musculoskeletal pain, alopecia and carpal tunnel syndrome. For each suspected ADR, a graphical exploratory analysis was performed and the WSP tool was applied for two censoring periods each. RESULTS: Both of the well-known and common ADRs (headache and musculoskeletal pain) were detected using the WSP tool, and the signals were present regardless of the censoring intervals used. A signal was also detected when the event was uncommon and the timing was likely to be an accurate reflection of onset time (alopecia). This signal was only present for some of the censoring intervals. As anticipated, no signals were raised in the control groups for these events regardless of the censoring interval used. The suspected ADR, which was uncommon and where reporting times may not reflect onset time accurately (carpal tunnel syndrome), was not detected. A signal was raised in the control group but its false-positive nature was visible in the exploratory graphical analysis, which led to it (frequent but for only a limited number of consecutive dates). CONCLUSION: This study illustrates the usability and examines the reliability of the WSP tool as a method for signal detection in electronic health records. When the events are uncommon the success of this method may depend on the reporting time accurately reflecting the true event onset time. The study has shown that further work is required to define the censoring periods. The addition of a control group is not required but may enhance causal inference by showing that other causes than the exposure may lead to a signal.
BACKGROUND: The WSP tool has previously been proposed as a method to detect signals for adverse drug reactions utilising time-to-event data without the need for a reference population. The aim of this study was to assess the performance of the tool on two well-known and two suspected adverse drug reactions for bisphosphonates that varied in both frequency and accuracy of reporting time. METHODS: The use of the WSP tool was investigated on data from a matched population cohort study involving data from UK primary care patients exposed to oral bisphosphonates. Four listed/suspected ADRs were selected for investigation: headache, musculoskeletal pain, alopecia and carpal tunnel syndrome. For each suspected ADR, a graphical exploratory analysis was performed and the WSP tool was applied for two censoring periods each. RESULTS: Both of the well-known and common ADRs (headache and musculoskeletal pain) were detected using the WSP tool, and the signals were present regardless of the censoring intervals used. A signal was also detected when the event was uncommon and the timing was likely to be an accurate reflection of onset time (alopecia). This signal was only present for some of the censoring intervals. As anticipated, no signals were raised in the control groups for these events regardless of the censoring interval used. The suspected ADR, which was uncommon and where reporting times may not reflect onset time accurately (carpal tunnel syndrome), was not detected. A signal was raised in the control group but its false-positive nature was visible in the exploratory graphical analysis, which led to it (frequent but for only a limited number of consecutive dates). CONCLUSION: This study illustrates the usability and examines the reliability of the WSP tool as a method for signal detection in electronic health records. When the events are uncommon the success of this method may depend on the reporting time accurately reflecting the true event onset time. The study has shown that further work is required to define the censoring periods. The addition of a control group is not required but may enhance causal inference by showing that other causes than the exposure may lead to a signal.
Authors: Preciosa M Coloma; Gianluca Trifirò; Martijn J Schuemie; Rosa Gini; Ron Herings; Julia Hippisley-Cox; Giampiero Mazzaglia; Gino Picelli; Giovanni Corrao; Lars Pedersen; Johan van der Lei; Miriam Sturkenboom Journal: Pharmacoepidemiol Drug Saf Date: 2012-02-08 Impact factor: 2.890
Authors: C A Naranjo; U Busto; E M Sellers; P Sandor; I Ruiz; E A Roberts; E Janecek; C Domecq; D J Greenblatt Journal: Clin Pharmacol Ther Date: 1981-08 Impact factor: 6.875
Authors: Saga Johansson; Mari-Ann Wallander; Francisco J de Abajo; Luis Alberto García Rodríguez Journal: Drug Saf Date: 2010-03-01 Impact factor: 5.606
Authors: Man Young Park; Dukyong Yoon; Kiyoung Lee; Seok Yun Kang; Inwhee Park; Suk-Hyang Lee; Woojae Kim; Hye Jin Kam; Young-Ho Lee; Ju Han Kim; Rae Woong Park Journal: Pharmacoepidemiol Drug Saf Date: 2011-06 Impact factor: 2.890
Authors: Martijn J Schuemie; Preciosa M Coloma; Huub Straatman; Ron M C Herings; Gianluca Trifirò; Justin Neil Matthews; David Prieto-Merino; Mariam Molokhia; Lars Pedersen; Rosa Gini; Francesco Innocenti; Giampiero Mazzaglia; Gino Picelli; Lorenza Scotti; Johan van der Lei; Miriam C J M Sturkenboom Journal: Med Care Date: 2012-10 Impact factor: 2.983