Clinical utility and perils of prescription drug monitoring program-based alert systems.
Clinical utility and perils of prescription drug monitoring program-based alert systems.
复制标题
基于处方药监测程序的警报系统的临床效用和风险。
DOI：
10.1097/j.pain.0000000000001157
复制
发表时间：
2018
期刊：
影响因子：
7.4
通讯作者：
Winterstein,AlmutG
中科院分区：
文献类型：
--
作者：
Hincapie-Castillo,JuanM;Wei,Yu-Jung;Winterstein,AlmutG
Furthermore, the proposed model has an extremely low positive predictive value. Even at the highest cutoff of 0.006 reported in Table 2, which captured roughly the upper decile of patients in the registry and only about one-third of overdose cases, positive predictive value is only 1.3%, resulting in 98/100 false positives. This is not only impractical for clinicians to act on, it also raises concerns about unintended consequences of mistakenly classifying patients. The “chilling effect” of the opioid abuse epidemic in the United States is steering away some prescribers from offering opioids as treatment option to their patients and an alert system such as the one proposed might compound this problem. 4 Rather than trying to optimize prediction with variables the authors found in the literature, it would be beneficial to explore why the model was not able to capture most cases at this threshold. A clinically meaningful model will likely require the integration of other relevant health information to improve prediction. Of note, the authors present a “parsimonious” model, which selected variables based on ease of measurement. Retention of variables after employment of formal statistical elimination techniques would assure readers that parsimony has indeed been achieved and follow best practices outlined in the TRIPOD statement for good quality predictive models. 2 Prediction models can offer significant utility if they achieve a clinically acceptable balance between false-positive and falsenegative cases and predict important clinical endpoints that warrant preventive intervention. Overdoses might not be the end, all be all for clinical decision support addressing opioid use. Although overdoses are rare and indeed undesirable, there might be other outcomes that could be predicted to avoid the manifestation of opioid use disorder (OUD) in the first place. Prevention of an imminent overdose seems to belong in the hands of addiction specialists and addressed by medicationassisted OUD treatment, where PDMP information may need a quite different interpretation than in the routine prescribing of opioids for pain.Measurement issues will continue to remain an obstacle for researchers in the field, and future efforts should be focused on identifying key markers in patients’ treatment trajectories. Such trajectories may greatly vary by opioid indication, patient comorbidities, age, life experiences, or other underexplored factors, emphasizing the need for careful construction of prediction models to mitigate erroneous flagging of patients who are not at risk of OUD and deserve adequate pain management, while missing those with OUD issues.
