This secondary analysis of the SELECT trial calculated NNT using the primary CV composite endpoint of the trial, in addition to two composite NNTs that included consideration of semaglutide’s effects on hospitalizations, and cardiovascular-, kidney-, or diabetes-related outcomes. In NNTCKM, we included subclinical events (e.g., HbA1c ≥ 6.5%, changes in eGFR) that, while not being hard outcomes, would be expected to form part of a clinician’s decision-making processes when reviewing therapy and which may then incur additional costs. The relative risk reductions associated with the broader NNTs (NNTEXTENDED, 20%; NNTCKM, 41%) were comparable to or greater than for the primary CV composite endpoint (20%). The 1- and 4-year NNTs were calculated to be 125 and 58 for NNT3P-MACE, 49 and 25 for NNTEXTENDED, and 20 and 11 for NNTCKM, respectively, indicating a broad benefit of semaglutide beyond the primary endpoint of SELECT.

The largest driver of difference between NNT3P-MACE and NNTEXTENDED is the inclusion of hospitalization events, as a function of their greater incidence compared with the primary CV composite. The largest driver of difference between NNTEXTENDED and NNTCKM is the inclusion of biochemical diabetes and the substantial reduction in rates of this event between placebo and semaglutide arms (see Table 1). Compared with NNT3P-MACE (based on the primary CV composite endpoint), the addition of all-cause hospitalization (ACH) in the time to first-event analysis reduces the counts of other fatal and non-fatal events, suggesting ACH is a precursor to more severe events. For example, MI event count in the semaglutide arm decreased from 239 to 168, suggesting that 71 patients in the semaglutide arm were hospitalized prior to experiencing an MI. Similar observations can be made for stroke and CV death. As a result of including ACH, the contribution of CV death and non-CV “other” death to NNTEXTENDED is relatively low. The decrease in NNT where ACH is incorporated indicates semaglutide’s efficacy in reducing the incidence of these earlier events with serious fatal and non-fatal sequelae. The observed reduction in NNT achieved by the inclusion of additional endpoints beyond the primary endpoint speaks to the limitations of a restricted definition of benefit. Patients and clinicians may be most interested in how a therapy can prevent fatal and non-fatal major CV events, and hence NNT3P-MACE. However, the prevention of (initially) less costly and less serious but more common outcomes, such as hospitalizations or progression to biochemical diabetes (as represented in NNTCKM), may be of equal relevance to health system administrators and payers seeking to manage limited resources in the face of the obesity epidemic.

NNT is specific to the results of a given comparison [25]; they are calculated from the absolute risk reduction between two treatment options in a single study, rather than from an absolute measure of clinical effect, and are sensitive to baseline risk, timeframe, and the outcomes considered. However, the reduction in composite NNTs observed in our analysis aligns with the broad beneficial effects on CV outcomes reported for semaglutide across indications [40]. A systematic review of eight CV outcome trials of GLP-1 RAs in patients with type 2 diabetes reported substantial variation of NNTs calculated to prevent one MACE, ranging from 83 to 429 at 1 year, 30 to 129 at 3 years, and 16 to 68 at 6 years [41]. Although the SELECT trial differs from the trials reviewed, most significantly in the recruitment of a population without diabetes (and hence potentially at lower baseline risk), it is encouraging that our calculated NNT3P-MACE falls within the ranges reported, contributing to the face-validity of our analysis.

The use of NNT as a tool to communicate value to stakeholders has advantages and disadvantages. NNTs are a simple measure that can provide valuable information alongside cost-effectiveness analyses when evaluating treatment benefits. They cannot replace the use of cost-effectiveness analyses [42], since they do not capture opportunity costs or healthcare costs, do not consider health-related quality of life benefits of treatment, are not readily comparable between trials [43], and cannot account for the costs and morbidity of multiple events experienced by individual patients. NNT are typically calculated based upon a low number of specific outcomes, e.g., primary trial endpoints, that are of clear significance to stakeholders. However, the standard approach to calculating NNT is unable to quantify the impact of competing risks where these exist. A simple resolution is to derive multiple NNTs with different events considered (hence, different competing risks) to capture the overall effect of a therapy [28].

The use of multiple, hierarchical NNTs addresses the issue of competing risks (e.g., non-CV death in NNT3P-MACE), while also recognizing the significance of a therapy’s broader treatment effects. There is already an appreciation within the medical literature that a narrow focus within outcomes trials misses the wider benefits of GLP-1 RAs across the CKM spectrum, and there are calls to review the design and analysis of outcomes trials in the CKM space to include more endpoints [44]. In this case, an approach to NNTs as reported here could become more common in the future to better communicate the total value of these drugs to healthcare systems.

A strength of this study is that our outcomes are derived from a major, pivotal clinical trial; the data underlying the NNT are specific, consistently measured, and are from a relevant clinical cohort across multiple countries and settings. Thus, the results generated should be robust, and well generalizable in this specific patient cohort. There have been six publications to date that consider the results of the SELECT trial, none of which calculate a NNT, either for the primary or composite endpoints [21, 34,35,36,37,38]. By presenting therapeutic effects on an absolute scale, NNTs help facilitate the practice of evidence-based medicine [45]. Furthermore, the hierarchy of NNTs presented in this study cannot be computed from the previous publications, owing to their dependence on the joint distribution of the individual components of the composite endpoints. Accordingly, our study takes a novel approach compared with the wider literature; the broad benefits of treatment with semaglutide are presented in a way that can inform clinical decision-makers, and which may have implications for the future of NNTs for multi-indication treatments.

Our study has limitations: the generalizability of the results of the SELECT trial to a real-world clinical population of patients with overweight or obesity and established CVD or to a population without established CVD is unknown, and the generalizability of our results is similarly uncertain. In addition, our assessment of value is limited by the design of outcome assessments in SELECT, in that any effect of semaglutide therapy on other obesity-related complications (e.g., sleep apnea, osteoarthritis, or atrial fibrillation) and their related healthcare resource use is uncertain. NNTs to prevent incidence of these outcomes cannot be calculated. We do not present subgroup analysis; it is likely that patients with comorbidities relevant to the outcomes assessed in our NNT models, such as those with prediabetes, would (all other things being equal) be associated with lower values of NNTs, which may have implications for clinical decision-making. Also, we recognize that presenting additional NNTs considering individual endpoints or groupings of endpoints (e.g., NNT5-point MACE, NNTMajor adverse kidney events) would be useful to some readers or would allow a degree of comparison between the results of different trials. In the current analysis, we have limited the number of NNTs presented to aid clarity. Finally, within the SELECT trial, the recording of a single HbA1c value ≥ 6.5% was defined as “progression to diabetes” [21]. There is uncertainty about whether patients reporting this would be expected to clinically present and be diagnosed with diabetes, as well as the extent to which the avoidance of such progression commensurately avoids all important clinical sequelae, and this may introduce a risk of overestimation in terms of “prevented diabetes.” However, as a subclinical outcome, the relevance and importance to patients of maintaining blood sugars below the hyperglycemic range to avoid a formal diabetes diagnosis is likely to mean its inclusion in our analyses is a valid approach to capturing value. This is particularly prescient, given the increasing awareness of how type 2 diabetes impacts on other components of CKM syndrome [13].