Weight changes after the last prescription of tirzepatide or semaglutide across a health network

GLP-1 receptor agonists (GLP-1RAs) such as semaglutide and tirzepatide have transformed obesity care and weight management, delivering unprecedented levels of weight loss in clinical trials []. For example, in the phase 3 STEP-1 trial, patients receiving once-weekly dosing of 2.4 mg semaglutide plus lifestyle intervention had an average of 15% weight loss over a 68-week period, compared to only 2.4% for patients receiving placebo and lifestyle intervention []. In the phase 3 SURMOUNT-1 trial, once-weekly treatment with 15 mg of tirzepatide plus lifestyle intervention resulted in 21% weight loss over a 72-week period, compared to 3% for placebo []. These significant impacts on weight management have been corroborated in analyses of real-world data, for example, in a comparative analysis of semaglutide and tirzepatide which showed average weight loss of 8% and 15% over a 1-year period, respectively []. Multiple randomized and real-world studies have also shown that treatment with semaglutide or tirzepatide more broadly improves cardiovascular health, including reductions in rates of cardiovascular events such as acute myocardial infarction and stroke []. ^{1 2 3 4 5–7}

Despite these benefits, real-world studies have shown that more than half of patients discontinue GLP-1RAs within one year of initiation []. There are various cited reasons for discontinuation, including gastrointestinal adverse effects, lack of treatment effect, out-of-pocket costs, and supply chain shortages []. Several studies have demonstrated reversal of weight loss and other cardiometabolic benefits after patients discontinue GLP-1RAs []. For example, in an extension of the STEP-1 trial, patients regained approximately two-thirds of their lost weight in the year following discontinuation of semaglutide, and cardiometabolic parameters such as blood pressure and hemoglobin A1c reverted to near pre-treatment levels over the same time period []. Similarly, in a post-hoc analysis of the SURMOUNT-4 trial, most patients regained weight after discontinuation of tirzepatide, and the degree of weight regain was associated with the degree of reversion of cardiometabolic parameters, including blood pressure, cholesterol levels, and hemoglobin A1c []. A systematic review of 18 randomized clinical trials came to similar conclusions regarding “metabolic rebound” following GLP-1RA discontinuation []. To date, it is not clear whether there are subsets of patients who are more likely to experience continued benefit after their therapeutic exposure. ^{8 9–11 12–14 13 14 12}

While the impact of discontinuation in a controlled setting (i.e. as part of a clinical trial protocol) is important to understand, it does not offer a comprehensive understanding of the true patient experience. Further, it fails to capture the perspective of the prescribing clinicians, whose patients may discontinue GLP-1RA therapy in line with or against their advice, transfer care to another provider within the same health system or be lost to follow-up for myriad reasons. This perspective has recently been even further complicated by the trend for patients to obtain GLP-1RA prescriptions from other sources, such as online providers or medical spas, with or without the knowledge of their physicians []. Thus, at the intersection of the patient and clinician experience, it is important to better understand weight trajectories following the last GLP-1RA prescription by a primary care or specialist provider. Here, we leverage the nSights platform to directly address this question using a large-scale longitudinal real-world dataset. ¹⁵

This study analyzed de-identified EHR data from a network of tertiary clinical centers tied to AMC in the USA through the nference nSights Analytics Platform. nference, in collaboration with AMC data partners, provided the de-identified data for this study. nference has established a secure data environment, hosted by and within each of the AMCs, that contains the AMC’s de-identified patient data. The provisioning of and access to this data are governed by an expert determination that satisfies the Health Insurance Portability and Accountability Act (HIPAA) Privacy Rule requirements for the de-identification of protected health information. Each AMC’s de-identified data environment is specifically designed and operated to enable access to and analysis of de-identified data without the need for Institutional Review Board (IRB) oversight, approval, or an exemption confirmation. Given these measures, informed consent and IRB review were not required for this study.

The code used in this study is not publicly available. Requests for additional information should be directed to the corresponding author.

Prior to analysis, all EHR data were de-identified under an expert determination consistent with the HIPAA Privacy Rule (45 CFR §164.514(b)(1)). The de-identification methodology employed a multi-layered transformation approach to both structured and unstructured data fields [,]. In structured data, direct identifiers including patient names and precise geographic locations were excluded entirely, while indirect identifiers underwent specific transformations: patient identifiers, medical record numbers, and accession numbers were replaced with one-way cryptographic hashes using confidential salts to preserve linkage across patient encounters; all dates were shifted backward by patient-specific random offsets (1–31 days) to preserve temporal relationships while obscuring exact event timing; the ZIP codes were truncated to two-digit state-level resolution; and continuous variables including age, height, weight, and body mass index were thresholded to prevent identification of extreme values (for example, ages ≥89 years transformed to ‘89+’ and BMI >40 transformed to ‘40+’). In unstructured clinical text, an ensemble de-identification system that combines attention-based deep learning models with rule-based methods achieved an estimated >99% recall for personally identifiable information detection, with detected identifiers replaced by plausible fictional surrogates []. ^{16 17 16}

To address heterogeneity in EHR data, we harmonized clinical variables, including medications, anthropometric measurements, and diagnoses, to standardized concepts. For medications, we first constructed a standardized drug concept database combining the nSights knowledge graph with RXNorm () hierarchies to capture ingredient, brand, and dose-specific information []. EHR medication records were matched using a hierarchical approach prioritizing RXNorm codes when available, followed by ingredient-level matching, and finally natural language processing and pattern matching on free-text medication orders when structured codes were absent. For anthropometric measurements (height, weight, BMI), we created a unified vocabulary from SNOMED (,) and LOINC () terminologies and matched EHR measurement descriptions using standardized text matching algorithms with abbreviation expansion and synonym resolution; ambiguous mappings were resolved using OpenAI GPT-4o () with summary statistics as context, followed by manual verification. For diagnoses, we developed a hierarchical disease concept database from the nSights knowledge graph and matched EHR diagnosis descriptions and codes by identifying the most specific common child concept in the hierarchy. This approach enabled consistent identification of clinical entities while preserving granularity where available. https://www.nlm.nih.gov/research/umls/rxnorm/index.html↗ https://www.snomed.org/↗ https://athena.ohdsi.org↗ https://loinc.org/↗ https://platform.openai.com/docs/models/gpt-4o↗ ¹⁸

There were 2567 patients in the semaglutide cohort and 1615 patients in the tirzepatide cohort. Demographic and baseline clinical characteristics are summarized in. Briefly, the semaglutide cohort was about 71.2% female and 77.3% white with a mean age of 55 years (SD: 13 years); the tirzepatide cohort was about 70.2% female, 85.3% white, and had a mean age of 55 years (SD: 12 years). At the time of the last prescription, the prevalence of type 2 diabetes and obesity, respectively, were 60.0% and 66.0% in the semaglutide cohort, and 77.0% and 82.5% in the tirzepatide cohort. The average baseline weight (i.e. approximately 12 months before the last prescription) in the semaglutide and tirzepatide cohorts was 103.3 and 108.5 kg, respectively, and average BMI was 34.6 and 35.5 kg/m, respectively. At least one prescription for a recommended maintenance dose (see the section “Methods”) during the year prior to the last prescription was present for 90.6% of patients taking semaglutide and 95.3% of patients taking tirzepatide. Table 1 2

Among these cohorts, we assessed the average percent change in weight in the year leading up to the last prescription (defined as month 0) and in the 6 months following the last prescription (). The patient-level weight change after the last GLP-1RA prescription is approximately normally distributed with a mean value of 0.29% (median value of 0), indicating that most patients did not experience significant weight gain following their last GLP-1RA prescription (). Patients were grouped retrospectively based on their post-last prescription trajectories into two clinically intuitive categories based on whether their weight increased by at least 2% in the 6 months following their last GLP-1RA prescription, labeled as “weight regain” and “non-regain” groups (see the section “Methods”). In both the tirzepatide- and semaglutide-treated cohorts, more than two-thirds of patients fell into the non-regain group (). Of note, the weight loss trends in the year leading up to the last prescription were quite similar between the weight regain and non-regain groups for both semaglutide (∼7% weight loss) and tirzepatide (∼11% weight loss). At the time of the last prescription, the weight and BMI distributions in the two groups were also similar for both medications (). Among patients who regained weight after their last GLP-1RA prescription (33.0% for semaglutide, 27.6% for tirzepatide), the average weight regain reached 4% within 4 months of the last prescription (). Fig. 1A Fig. 1B Fig. 2 Fig,  3 Fig. 2

We next evaluated whether lifestyle factors, specifically diet and physical activity, might be associated with sustained weight maintenance. At a representative site, we applied LLMs to analyze the frequency of lifestyle counseling documented in longitudinal EHR notes (see the section “Methods”) among patients who did not experience weight regain (= 226) versus those who did (= 74). Patients without weight regain were more likely to have exercise counseling mentions than those with regain (26.2% vs. 14.7%;= .04;). In contrast, rates of diet counseling after the last prescription were similar between these groups (22.3% vs. 23.5%;= .83;). N N P P Fig. 4 Fig. 4

The analyses presented thus far are anchored on the date of each patient’s last prescription, which does not necessarily correspond to medication discontinuation. To better understand the relationship between these events, we deployed an LLM across clinical notes from a representative subset of 300 patients (159 taking semaglutide, 141 taking tirzepatide) to determine how frequently there was supporting clinician documentation of treatment cessation within ±90 days of the last prescription (see the section “Methods”). There was such documentation for 119 patients (40%) overall, including 68 of 159 (43%) semaglutide patients and 61 of 141 (36%) tirzepatide patients (). Among the 119 patients with clinician-documented GLP-1RA discontinuation, 86 (72%) were in the non-regain group, including 51 of 68 (75%) semaglutide patients and 35 of 51 (69%) tirzepatide patients (). It was rare to find notes indicating that patients subsequently obtained the same or another GLP-1RA from another source (), although this likely underestimates the true prevalence of such behavior due to incomplete reporting and documentation. Table 2 Table 2 Table 2

Overall, this study highlights that while patients with an active GLP-1RA prescription tend to lose significant weight, weight regain is not universally observed after the last prescription captured in the EHR. There are likely multiple factors contributing to this observation. For example, as described above, exercise counseling was documented more frequently in the non-regain cohorts. This suggests that lifestyle factors such as exercise may play an important role in modulating weight loss durability following the last GLP-1RA prescription and motivates randomized clinical trials to directly test this hypothesis. Further, some patients may have continued to have treatment exposure following the date of the last prescription, whether related to this prescription itself or medication access from another source, such as a medical spa, telehealth provider, and/or compounding pharmacy. In any case, the findings of this study can help both patients and clinicians better understand what happens in the real world after a primary care or specialist physician stops prescribing a GLP-1RA.

For multiple reasons, the data presented here is not contradictory to prior randomized controlled trials (RCTs) that have demonstrated weight regain after discontinuation of GLP-1RAs [,,,]. First, this analysis focuses on weight trajectories that follow the last GLP-1RA prescription by a primary care or specialist physician, a unique “real-world index event” that has not been explored in previous studies. While our follow-up analysis on patients with confirmed discontinuation in clinical notes yielded similar findings, it is important to note that assessment of discontinuation via EHR notes has its own challenges and limitations, including incomplete documentation and automated inclusion of structured content in notes (e.g. medication lists). Second, RCTs have measured weight rebound under controlled withdrawal settings (e.g. withdrawal at a specified timepoint after therapy initiation), whereas real-world data reflect a combination of pharmacology, adaptive human behavior, selective discontinuation, flexible re-initiation patterns, and more. Indeed, real-world treatment cessation is heterogenous and often partial, staggered, or followed by re-initiation, dose tapering, lifestyle changes, or adjunctive therapies, all of which could blunt or magnify metabolic rebound physiology [,]. Patients who discontinue therapy and do not restart within the following 6 months, as is the case for the cohorts analyzed here, are by definition a selected subset who likely have unique characteristics relating to treatment exposures and behavioral adaptations. The implication of this study is not that the risk of weight rebound is negligible but rather that durable response after prescription discontinuation may be achievable with an appropriate framework in place. Accordingly, it may be useful to develop systems that help clinicians better identify who can safely discontinue, who requires continued maintenance therapy, and when sustained intervention is indicated to preserve metabolic benefits. ^{12 14 19 20 8 21}

This study has several limitations. First, it is a retrospective analysis of real-world data and is thus prone to multiple sources of confounding. For example, the inclusion criteria applied in this study (i.e. multiple prescriptions for the same medication and multiple weight measurements over a 11.5-year-period) could lead to selection bias with enrichment for patients who are more likely to have lost and kept off weight. Further, medical comorbidities and concomitant medications could impact treatment response and post-treatment weight maintenance patterns. For example, several anti-diabetic medications are known to promote weight increase (e.g. insulin) or decrease (e.g. metformin) []. Second, the follow-up interval is limited to 6 months, whereas prior studies have explored longer durations following treatment cessation [,,,]. The shorter-interval follow-up could lead to underestimation of patients who regain weight after the last prescription, and the analyses presented here should be interpreted as early trajectories following prescription discontinuation. Indeed, in the non-regain cohorts described here, there is an upward deflection of weight at the 6-month timepoint (), which could in part reflect impending weight regain for at least a subset of the cohort. Conversely, re-initiation of therapy after the 6-month observation window assessed in this study would likely impact longer-term weight trajectories. Studies with longer follow-up times will be important to fully characterize weight regain dynamics. Third, there is certainly variability in treatment adherence, even leading up to the last prescription. To enrich for patients who were more likely to have had true therapy exposure, we required patients to have at least three GLP-1RA prescriptions at most 4 months apart during a one-year period in the participating health systems for inclusion in the study. However, it is likely that some of these patients missed doses or stopped taking the medication before the last documented prescription. Conversely, as mentioned previously, it is likely that some patients continued to receive the medication after their last documented prescription or after the observation window from other sources that are not captured in the EHR, such as telehealth providers or compounding pharmacies. This is an important potential source of misclassification bias that could influence the observed weight trajectories. Finally, this study did not account for social determinants of health that impact access to therapies, prescription fill rates, and treatment adherence, including but not limited to income and insurance status []. Future analyses are warranted to explore how these factors impact heterogeneity in weight maintenance in the post-GLP-1RA prescription interval. ^{22 13 14 19 23} Fig. 2 ²⁴

In conclusion, in the study population presented here, weight loss after GLP-1RA therapy did not uniformly reverse after the last EHR-documented prescriptions. Rather, a majority of patients maintained their weight or continued to lose weight, while a smaller but important subset experienced weight regain. These findings illustrate the importance of a nuanced view of long-term weight regulation in the era of GLP-1 therapies and motivate future multimodal monitoring of patients after prescription discontinuation and/or therapy cessation which integrate biology, behavior, comorbidities, and drug dosing information. Such investigations could unlock features that can help to predict patients who are likely to show sustained benefit versus metabolic rebound.

KM, GV, AJV and VS are employees of nference, inc., which conducts research collaborations with various biopharmaceutical companies, including AstraZeneca, Eli Lilly and Company, and Novo Nordisk A/S, whose GLP-1 receptor agonist products (semaglutide and tirzepatide formulations) are included in this study. None of these companies, nor any other nference collaborator, funded, supported, or had any role in the independent study design, data acquisition, analysis, interpretation, manuscript preparation, or the decision to submit this work for publication. All analyses were conducted by the authors using de-identified EHR data. CMG reports consulting with nference. The authors declare no additional competing interests.