This retrospective case-control investigation compared two opioid risk mitigation tools used by pharmacists and clinical teams to identify hospitalized patients at highest risk for opioid-induced respiratory depression (OIRD). The explicit aim was to evaluate the capacity of the Michigan Opioid Safety Score (MOSS) and the PRediction of Opioid-induced Respiratory Depression In patients monitored by capnoGraphY (PRODIGY) tool to correctly identify nonsurgical, noncritical adult inpatients who experienced OIRD while receiving opioids during their hospitalization.
The study employed a retrospective case-control design. Manual calculation of MOSS and PRODIGY scores was performed for included patients. Key exclusion criteria were: admission to the emergency department, maternity ward, or medical, cardiac, or neurology intensive care units; no opioid administration during hospitalization; length of stay less than 24 hours; and observation-status admissions. The focus was on nonsurgical, noncritical adult inpatients exposed to opioids.
The two tools compared were the Michigan Opioid Safety Score (MOSS) and the PRODIGY risk tool. Both are intended to support identification of patients at risk for opioid-related respiratory compromise, enabling targeted monitoring or interventions. The scores for each patient were calculated manually from available clinical data for the study cohort.
Primary performance measures reported were sensitivity and specificity for detecting OIRD events. Receiver operating characteristic (ROC) curves and area under the curve (AUC) values were also calculated to describe overall discriminative performance of each tool. The study reported between-tool comparisons, confidence intervals, and P values for differences in sensitivity and specificity.
Sample-size planning indicated that 388 patients were required to establish statistical power for the analysis. A total of 778 patients were initially selected for the study dataset; after applying inclusion and exclusion criteria, 216 patients met the criteria and were included in the analysis.
PRODIGY demonstrated a sensitivity of 86.62% and a specificity of 20.27% for predicting OIRD events in this cohort. By contrast, MOSS demonstrated a sensitivity of 2.82% and a specificity of 100%.
Statistical comparison between the tools indicated a significantly greater sensitivity for PRODIGY compared with MOSS. The reported difference in sensitivity was 83.8% (95% CI, 77.4%–90.2%; P < 0.0001), supporting PRODIGY as the more sensitive tool for detecting OIRD. Conversely, there was a statistically significant difference in specificity favoring MOSS; the reported difference in specificity was −79.7% (95% CI, −88.9% to −70.6%; P < 0.0001), indicating MOSS classified far fewer false positives in this sample but missed the majority of OIRD events.
ROC analysis produced an AUC of 0.59 for PRODIGY and 0.48 for MOSS. An AUC of 0.59 for PRODIGY indicates modest discriminative ability in this dataset, whereas an AUC of 0.48 for MOSS indicates performance no better than chance in overall discrimination of patients who developed OIRD within the studied population.
The authors concluded that PRODIGY is more sensitive than MOSS for identifying risk of opioid-induced respiratory depression among nonsurgical, noncritical hospitalized adults who receive opioids. MOSS demonstrated perfect specificity in this cohort but very low sensitivity, whereas PRODIGY identified a substantially larger proportion of OIRD events despite low specificity. The results imply a trade-off between sensitivity and specificity when selecting a risk tool: PRODIGY may be preferable where detection of most at-risk patients is prioritized, while MOSS may be useful where false positives must be minimized.
All numerical results, sample counts, inclusion and exclusion criteria, and statistical outcomes reported here are drawn from the PubMed abstract and article citation (Am J Health Syst Pharm. 2026 Aug 19;83(17):e683-e690; PMID 41208413; DOI 10.1093/ajhp/zxaf307). The abstract provides the primary performance metrics, sample counts, and conclusions. Additional methodological details, subgroup analyses, demographic breakdowns, or operational definitions beyond what appears in the abstract were not reported in the source provided here.