Comparative effectiveness of long-term anti-obesity medications in routine clinical practice: a retrospective cohort study using real-world data Authors: Chul-Gyu Kim, Jung-Min Koh, Kyun-Seop Bae Published: 2025 ABSTRACT AbstractOrlistat, phentermine/topiramate, and liraglutide are widely used obesity pharmacotherapies, but comparative data from Asian real-world settings remain limited. We compared their effects on body weight, glycemic control, and persistence. We performed a retrospective cohort study using electronic health records from a tertiary hospital in Korea. Adults prescribed orlistat, phentermine/topiramate, or liraglutide between 2018 and 2025 were included. The primary endpoint was percentage weight change at 6 months. Secondary endpoints were the proportions achieving ≥5% and ≥10% weight loss, weight change at 3, 6, and 12 months, change in HbA1c, and 12-month treatment persistence. We applied multivariable linear and logistic regression, linear mixed-effects models, propensity score-weighted models, and Kaplan-Meier and restricted mean survival time (RMST) methods. Among 1,910 treatment episodes, 505 involved orlistat, 777 phentermine/topiramate, and 628 liraglutide. Phentermine/topiramate showed greater mean percentage weight loss at 6 months than orlistat (adjusted difference, −1.59 percentage points; p < 0.001), whereas liraglutide produced similar weight loss to orlistat. The odds of achieving ≥5% weight loss were higher with phentermine/topiramate than with orlistat (adjusted odds ratio, 2.21), while liraglutide showed a smaller effect. Liraglutide tended to reduce HbA1c more than orlistat, but this difference was attenuated after propensity score weighting. Twelve-month treatment persistence and RMST were lower with liraglutide than with the oral agents. In this real-world cohort, phentermine/topiramate achieved the most pronounced and sustained weight loss. Liraglutide and orlistat produced similar overall weight reduction, but liraglutide showed lower long-term persistence.Keywords: Obesity; Anti-Obesity Agents; Weight Loss, Electronic Health Records INTRODUCTION Obesity has emerged as a major public health problem in Korea. Over the past decade, the prevalence of obesity among adult men has increased steadily, whereas the prevalence among women has remained relatively stable [1]. Obesity is a chronic disease and an established risk factor for hypertension, type 2 diabetes mellitus, dyslipidemia, cardiovascular disease, and certain cancers [2], and more recently for severe outcomes of coronavirus disease 2019 infection [3]. Effective long-term weight management is therefore crucial for reducing the burden of obesity-related complications. Lifestyle modification with dietary intervention and physical activity remains the cornerstone of obesity management. However, in adults with a body mass index (BMI) of 25 kg/m2 or higher who fail to achieve weight loss with non-pharmacologic therapy, clinical practice guidelines recommend adding pharmacologic therapy [4]. In Korea, several long-term anti-obesity medications have been approved, including orlistat, phentermine/topiramate, naltrexone/bupropion, and liraglutide 3.0 mg, whereas lorcaserin was withdrawn because of concerns about cancer risk [4,5]. Among these agents, orlistat, phentermine/topiramate, and liraglutide have been widely used in routine care. Randomized controlled trials have demonstrated that each of these agents produces clinically meaningful weight loss, and some have shown additional benefits on cardiometabolic risk factors. Phentermine/topiramate has consistently produced large weight reductions among approved long-term agents [4,6,7], whereas liraglutide, a glucagon-like peptide-1 receptor agonist, has demonstrated beneficial effects on glycemic control in patients with type 2 diabetes [8,9,10]. However, trial populations are highly selected and intensively managed, limiting the generalizability of trial findings to routine care. Real-world data are increasingly recognized as an important complement to trial data, providing insights into effectiveness, safety, and treatment persistence in broader, more heterogeneous patient populations [11,12]. Several real-world studies have evaluated individual anti-obesity medications or compared limited subsets of agents. A Korean single-center study including multiple anti-obesity drugs reported significant reductions in body weight and body fat mass but did not evaluate detailed cardiometabolic parameters [13]. In a real-world primary care cohort, Leventhal-Perek et al. [14] compared liraglutide 3 mg, lorcaserin, and orlistat and found clinically meaningful weight loss and relatively favorable treatment persistence with liraglutide. A systematic review of real-world outcomes reported that clinically meaningful weight loss was generally observed with liraglutide 3 mg, semaglutide 2.4 mg, orlistat, naltrexone/bupropion, and phentermine/topiramate, but improvements in blood pressure, lipids, and glycemic measures were limited or inconsistent [15]. To date, there is a paucity of real-world comparative data on long-term anti-obesity pharmacotherapies in Korean clinical practice, particularly regarding head-to-head comparisons of orlistat, phentermine/topiramate, and liraglutide with respect to weight loss, cardiometabolic parameters, and treatment persistence. Understanding how these drugs perform in routine care, beyond trial settings, is essential for informing individualized treatment decisions and optimizing obesity management strategies. The present study leveraged electronic health record (EHR) data from a large tertiary hospital to compare the real-world effectiveness of orlistat, phentermine/topiramate, and liraglutide. Specifically, we aimed to 1) compare short- and intermediate-term changes in body weight and metabolic parameters across the three drugs; 2) evaluate the proportions of patients achieving clinically meaningful weight loss; and 3) assess 12-month treatment persistence. We further used propensity score–based methods to mitigate confounding by indication and explore the robustness of our findings. METHODS Study design and data source We conducted a retrospective cohort study using data from the Asan BiomedicaL research Environment (ABLE), the research data extraction system of Asan Medical Center, a large tertiary care hospital in Seoul, Korea. ABLE provides de-identified clinical data derived from EHRs, including demographic information, diagnoses, prescriptions, vital signs, and laboratory results. The study period spanned from 1 January 2018 to 31 August 2025. All data processing and analyses were performed within the secure institutional environment using de-identified datasets. Study population and treatment episodes We identified adult patients (≥ 18 years) who received at least one prescription for orlistat, phentermine/topiramate, or liraglutide during the study period. For each patient and drug, we defined a treatment episode as the interval from the first prescription date (index date) to the last prescription date for that specific agent, before a gap exceeding a predefined grace period or switching to a different long-term anti-obesity medication. The expected end date of each prescription was calculated from the prescribed days of supply. Treatment discontinuation was defined as a gap of more than 60 days between the expected end date of a prescription and the next prescription date. The grace period of 60 days was chosen to reflect irregular clinic visits and refill patterns in real-world practice and to avoid misclassifying short delays in refilling as permanent discontinuation. Key inclusion criteria for the analytic cohort were: 1) at least one prescription for orlistat, phentermine/topiramate, or liraglutide between 1 January 2018 and 31 August 2025; 2) availability of baseline body weight within a prespecified window around the index date; and 3) availability of follow-up body weight at approximately 6 months. Patients with missing general chemistry results were excluded from analyses involving laboratory outcomes. Patients could contribute more than one treatment episode if they received different agents at different times, and each episode was analyzed separately. Exposure: anti-obesity medications The exposure of interest was treatment with one of the following long-term anti-obesity medications: orlistat, phentermine/topiramate, or liraglutide 3.0 mg (obesity dose). Drug identity and prescription dates were obtained from EHR prescription records. In Korea, phentermine/topiramate is available as a fixed-dose extended-release combination product. In this study, phentermine/topiramate episodes were identified using product codes for the fixed-dose combination, according to the coding structure in our database. Outcomes The primary outcome was percentage change in body weight from baseline to 6 months, calculated as with negative values indicating weight loss. Baseline weight was defined as the body weight measurement closest to the index date within a prespecified window around the index date, usually at the same visit or a visit within several weeks. For each nominal follow-up time point (3, 6, and 12 months), we calculated a target date as the index date plus 90, 180, or 365 days, respectively, and then selected the weight measurement closest to that target date within a priori allowable interval that reflected typical clinic follow-up patterns. When more than one measurement fell within the interval, we used the value closest in time to the target date. If no measurement was available within the allowable interval, the episode did not contribute to analyses for that time point. Because weights were obtained as part of routine care rather than at protocol-mandated visits, the exact timing of measurements varied across patients and some time-window bias is possible, as discussed below. Secondary outcomes included: 1) The proportion of episodes achieving ≥ 5% and ≥ 10% weight loss at 6 months. 1) The proportion of episodes achieving ≥ 5% and ≥ 10% weight loss at 6 months. 2) Percentage weight change at approximately 3, 6, and 12 months. 2) Percentage weight change at approximately 3, 6, and 12 months. 3) Change in glycated hemoglobin (HbA1c) from baseline to 6 months among episodes with available measurements. 3) Change in glycated hemoglobin (HbA1c) from baseline to 6 months among episodes with available measurements. 4) Time from index date to treatment discontinuation within 12 months. 4) Time from index date to treatment discontinuation within 12 months. Covariates Baseline covariates were assessed at or before the index date and included age, sex, baseline BMI, and comorbidities (diabetes mellitus, hypertension, dyslipidemia) identified from diagnosis codes. Baseline laboratory values, including HbA1c and metabolic parameters, were also extracted where available. Statistical analysis All analyses were conducted using R version 4.5.2 (R Foundation for Statistical Computing, Vienna, Austria). Baseline characteristics were summarized by drug group using means and standard deviations for continuous variables and frequencies and percentages for categorical variables. Differences among the three groups were explored using analysis of variance or Kruskal–Wallis tests for continuous variables and χ2 tests for categorical variables. To quantify baseline imbalance across the three treatment groups, we calculated standardized mean differences (SMDs) for each covariate. For continuous covariates, we used a global SMD based on the one-way analysis of variance effect size (Cohen’s f), defined as the square root of the between-group variance divided by the pooled within-group variance. For binary covariates, conventional pairwise SMDs were computed using pooled standard deviations of the underlying proportions. Values around 0.2 to 0.3 were interpreted as small to moderate imbalance. These SMDs are reported in Table 1. Table 1Baseline characteristics by drug groupVariableDrugOrlistatP/TLiraglutideSMD*No.505777628Age (yr)(mean ± SD)44.07 ± 17.0445.49 ± 12.4139.45 ± 15.030.273BMI (kg/m2)(mean ± SD)31.47 ± 5.2631.75 ± 5.6633.25 ± 6.290.204Weight (kg)(mean ± SD)85.0 ± 19.5385.65 ± 19.2591.22 ± 22.190.199HbA1c (%)(mean ± SD)5.86 ± 0.876.30 ± 1.176.41 ± 1.400.328Male(%)32.924.332.60.127ComorbidityDM(%)11.324.323.60.231Hypertension(%)18.622.824.20.091Dyslipidemia(%)35.639.138.20.048P/T, phentermine/topiramate; SMD, standardized mean difference; SD, standard deviation; BMI, body mass index; HbA1c, glycated hemoglobin; DM, type 2 diabetes mellitus.*For continuous covariates with three treatment groups, we computed a global SMD using a one-way analysis of variance effect size (Cohen’s f), defined as the square root of the weighted between-group variance divided by the pooled within-group variance. Baseline characteristics by drug group P/T, phentermine/topiramate; SMD, standardized mean difference; SD, standard deviation; BMI, body mass index; HbA1c, glycated hemoglobin; DM, type 2 diabetes mellitus. *For continuous covariates with three treatment groups, we computed a global SMD using a one-way analysis of variance effect size (Cohen’s f), defined as the square root of the weighted between-group variance divided by the pooled within-group variance. For the primary endpoint, multivariable linear regression models were fitted with drug group as the main independent variable, using orlistat as the reference, and adjusting for age, sex, baseline BMI, and comorbidities. For binary endpoints (achievement of at least 5% and at least 10% weight loss at 6 months), we fitted multivariable logistic regression models and reported odds ratios (ORs) with 95% confidence intervals (CIs), using orlistat as the reference group. Response was coded as 1 if the percentage weight change was ≤ −5% or ≤ −10%, respectively, and 0 otherwise. For longitudinal weight trajectories, we used linear mixed-effects models with percentage weight change at 3, 6, and 12 months as the dependent variable. Fixed effects included time (categorical; 3, 6, and 12 months), drug group, the time-by-drug interaction, age at index, sex, and baseline BMI. A random intercept for each subject was included to account for within-subject correlation. Adjusted marginal means and pairwise differences between drug groups at each time point were obtained using the emmeans package, with Tukey adjustment for multiple testing across the three treatments and three time points. To address confounding by indication, we performed pairwise propensity score—weighted analyses for orlistat versus phentermine/topiramate and orlistat versus liraglutide. Propensity scores were estimated by logistic regression including age, sex, baseline BMI, comorbidities (diabetes mellitus, hypertension, dyslipidemia), and baseline HbA1c for glycemic outcomes. Inverse probability of treatment weights for the average treatment effect were computed, and weighted linear and logistic regression models were used to estimate differences in percentage weight change and odds of achieving at least 5% and at least 10% weight loss. For HbA1c, propensity score—weighted linear regression was used. Treatment persistence was analyzed using Kaplan–Meier methods, with the event defined as treatment discontinuation within 365 days of the index date. Time was measured from the index date to the earliest of treatment discontinuation, 365 days, or the end of available follow-up. Episodes without a recorded discontinuation before 365 days were censored at the time of last expected prescription coverage or the administrative end of follow-up. Survival probabilities at 180 and 365 days were estimated for each drug. In addition, restricted mean survival time (RMST) up to 365 days was computed for each drug using the survRM2 package, and pairwise RMST differences and ratios were calculated. Exploratory Cox proportional hazards models were also fitted, but because the proportional hazards assumption was strongly violated in cox.zph diagnostics, we relied primarily on Kaplan–Meier and RMST-based summaries for interpretation. All tests were 2-sided, and p-values < 0.05 were considered statistically significant. Ethical considerations This study was conducted in accordance with the Declaration of Helsinki and relevant Korean regulations, including the Personal Information Protection Act, the Medical Service Act, and the Bioethics and Safety Act. The study protocol was reviewed and approved by the Institutional Review Board of Asan Medical Center (approval No. 2024-1240). Because the study used de-identified data from completed clinical encounters without direct patient contact, the requirement for informed consent was waived. RESULTS Baseline characteristics A total of 1,910 treatment episodes were included in the primary analysis: 505 orlistat, 777 phentermine/topiramate, and 628 liraglutide episodes (Table 1). Some patients contributed more than one episode if they received different agents at different times. Baseline characteristics differed across drug groups. On average, patients receiving liraglutide were younger (mean age 39.5 years) and had higher BMI (mean 33.3 kg/m2) than those receiving orlistat or phentermine/topiramate (BMI 31.5 and 31.8 kg/m2, respectively). The prevalence of type 2 diabetes mellitus was 11.3% in the orlistat group, 24.3% in the phentermine/topiramate group, and 23.6% in the liraglutide group; hypertension and dyslipidemia were also more frequent in the phentermine/topiramate and liraglutide groups than in the orlistat group (Table 1). These patterns suggest preferential use of phentermine/topiramate and liraglutide in patients with more severe obesity and cardiometabolic risk. Mean baseline weight and HbA1c were also higher in the phentermine/topiramate and liraglutide groups than in the orlistat group (Table 1). Global SMDs for continuous covariates ranged from approximately 0.20 for BMI and weight to 0.33 for HbA1c, indicating small to moderate baseline imbalance across treatment groups. Weight change at 6 months In multivariable linear regression adjusting for age, sex, baseline BMI, and comorbidities, phentermine/topiramate was associated with significantly greater percentage weight loss at 6 months compared with orlistat (difference, −1.59 percentage points, standard error [SE], 0.35; p < 0.001). Liraglutide showed a similar magnitude of weight reduction compared with orlistat; the adjusted difference was small (0.21 percentage points, SE, 0.39) and not statistically significant (p = 0.58) (Table 2). Table 2Six-month weight change and weight-loss response by drug group: multivariable and propensity score–weighted analysesVariableEstimateSEp-valueContinuous outcome: % weight change at 6 monIntercept (orlistat based)1.8881.0440.07Phentermine/Topiramate−1.5880.347< 0.01Liraglutide0.2140.3870.58Age (yr)−0.0140.0100.17Sex (male)1.1750.338< 0.01BMI (kg/m2)−0.2060.027< 0.01DM1.0090.382< 0.01Hypertension0.6450.3820.09Dyslipidemia−0.2210.3160.48Binary outcome: ≥ 5% weight loss at 6 monIntercept (orlistat based)−1.6030.355< 0.01Phentermine/Topiramate0.7940.120< 0.01Liraglutide0.1960.1250.12Age (yr)−0.0020.0030.55Sex (male)−0.4880.111< 0.01BMI (kg/m2)0.0530.009< 0.01DM−0.4370.125< 0.01Hypertension−0.2830.1250.02Dyslipidemia−0.0230.1030.83Binary outcome: ≥ 10% weight loss at 6 monIntercept (orlistat based)−2.5110.409< 0.01Phentermine/Topiramate0.3120.1450.03Liraglutide−0.3670.1630.02Age (yr)−0.0120.004< 0.01Sex (male)−0.4120.140< 0.01BMI (kg/m2)0.0550.010< 0.01DM−0.4420.166< 0.01Hypertension−0.2230.1600.15Dyslipidemia0.1130.1270.37SE, standard error; BMI, body mass index; DM, type 2 diabetes mellitus. Six-month weight change and weight-loss response by drug group: multivariable and propensity score–weighted analyses SE, standard error; BMI, body mass index; DM, type 2 diabetes mellitus. For the binary endpoint of ≥ 5% weight loss at 6 months, the adjusted odds of response were significantly higher with phentermine/topiramate than with orlistat (OR, 2.21; 95% CI, 1.75–2.80), while liraglutide showed a smaller, non-significant increase in the odds of response compared with orlistat (OR, 1.22; 95% CI, 0.95–1.56) (Table 3). In propensity score—weighted analyses, the superiority of phentermine/topiramate over orlistat persisted, with a similar difference in mean percentage weight loss (−1.61 ± 0.38 percentage points) and a slightly attenuated but still elevated odds of achieving ≥ 5% weight loss (OR, 2.11; 95% CI, 1.80–2.47). For the liraglutide vs orlistat comparison, differences in mean percentage weight loss remained small and non-significant after weighting, but liraglutide was associated with a modest yet statistically significant increase in the odds of achieving ≥ 5% weight loss (OR, 1.19; 95% CI, 1.01–1.40) (Table 4). For the more stringent endpoint of at least 10% weight loss at 6 months, adjusted logistic regression showed that phentermine/topiramate was associated with higher odds of response than orlistat (OR, 1.37; 95% CI, 1.03–1.82), whereas liraglutide was associated with lower odds of achieving at least 10% weight loss compared with orlistat (OR, 0.69; 95% CI, 0.50–0.95) (Table 3). Table 3Six-month weight change and weight-loss response by drug group: multivariable and propensity score–weighted analyses (odds ratio by multiple logistic regression)Drug> 5% body weight change> 10% body weight changeOdds ratio (SE)95% CIOdds ratio (SE)95% CIP/T - orlistat2.211.75–2.801.371.03–1.82Lira - orlistat1.220.95–1.560.6930.504–0.953SE, standard error; CI, confidence interval; P/T, phentermine/topiramate; Lira, liraglutide. Six-month weight change and weight-loss response by drug group: multivariable and propensity score–weighted analyses (odds ratio by multiple logistic regression) SE, standard error; CI, confidence interval; P/T, phentermine/topiramate; Lira, liraglutide. Table 4Six-month weight change and weight-loss response by drug group: multivariable and propensity score–weighted analyses (propensity score adjusted pairwise difference and odds ratio)DrugΔ% weight at 6 mon (mean ± SE)> 5% body weight change> 10% body weight changeOdds ratio95% CIOdds ratio95% CIP/T - orlistat−1.609 ± 0.3812.111.80–2.471.311.09–1.59Lira - orlistat0.212 ± 0.3591.191.01–1.400.6580.530–0.815SE, standard error; CI, confidence interval. Six-month weight change and weight-loss response by drug group: multivariable and propensity score–weighted analyses (propensity score adjusted pairwise difference and odds ratio) SE, standard error; CI, confidence interval. Fig. 1 shows adjusted trajectories of percentage weight change at 3, 6, and 12 months from the linear mixed-effects models. Fig. 2 illustrates the distribution of percentage weight change at 6 months in each drug group. Adjusted trajectories of percentage weight change at 3, 6, and 12 months by anti-obesity medication. Lines represent estimated marginal means from linear mixed-effects models adjusted for age, sex, baseline body mass index, and comorbidities, with 95% confidence intervals. Boxplots of percentage weight change at 6 months by drug group. Boxplots display median, interquartile range, and outliers for percentage weight change at 6 months for each anti-obesity medication. Horizontal dashed lines indicate −5% and −10% thresholds. Longitudinal weight trajectories Linear mixed-effects models incorporating repeated measures of percentage weight change at 3, 6, and 12 months showed that all three drugs produced early weight loss that was largely maintained up to 12 months (Table 5, Fig. 1). The numbers of episodes contributing to each time point for each drug are reported in Table 5. Adjusted marginal means from these models indicated that phentermine/topiramate consistently achieved the greatest percentage weight loss at each time point. At 6 months, the estimated mean weight change was −4.29% (SE, 0.23) in the orlistat group, −5.40% (SE, 0.19) in the phentermine/topiramate group, and −4.24% (SE, 0.21) in the liraglutide group (Table 5). Pairwise comparisons demonstrated that orlistat was associated with less weight loss than phentermine/topiramate at all time points (for example, orlistat minus phentermine/topiramate 1.11 ± 0.29 percentage points at 6 months; p < 0.01), whereas differences between orlistat and liraglutide were small and not statistically significant (Table 6). Table 5Adjusted mean percentage weight change at 3, 6, and 12 months by drug group from linear mixed-effects models (estimated marginal means)Drug3 mon6 mon12 monMean ± SENo.Mean ± SENo.Mean ± SENo.Orlistat−2.94 ± 0.1671,058−4.29 ± 0.229505−4.28 ± 0.305265Phentermine/Topiramate−3.54 ± 0.1541,323−5.40 ± 0.191777−5.62 ± 0.247421Liraglutide−3.16 ± 0.1441,523−4.24 ± 0.208628−4.03 ± 0.296284Estimates are from a linear mixed-effects model with percent weight change as the dependent variable and fixed effects for time (3, 6, and 12 months), drug, the time × drug interaction, age at index, sex, and baseline body mass index. A random intercept for each subject was included.SE, standard error. Adjusted mean percentage weight change at 3, 6, and 12 months by drug group from linear mixed-effects models (estimated marginal means) Estimates are from a linear mixed-effects model with percent weight change as the dependent variable and fixed effects for time (3, 6, and 12 months), drug, the time × drug interaction, age at index, sex, and baseline body mass index. A random intercept for each subject was included. SE, standard error. Table 6Adjusted mean percentage weight change at 3, 6, and 12 months by drug group from linear mixed-effects models (pairwise difference)Drug3 mon6 mon12 monMean ± SEp-valueMean ± SEp-valueMean ± SEp-valueOrlistat - P/T0.602 ± 0.2130.011.112 ± 0.286< 0.011.338 ± 0.382< 0.01Orlistat - Liraglutide0.218 ± 0.2070.55−0.050 ± 0.2970.98−0.244 ± 0.4160.83P/T - Liraglutide−0.385 ± 0.1950.12−1.162 ± 0.270< 0.01−1.582 ± 0.376< 0.01Pairwise comparisons between drugs at each time point were obtained from the mixed-effects model using Tukey adjustment for multiple testing.SE, standard error; P/T: phentermine/topiramate. Adjusted mean percentage weight change at 3, 6, and 12 months by drug group from linear mixed-effects models (pairwise difference) Pairwise comparisons between drugs at each time point were obtained from the mixed-effects model using Tukey adjustment for multiple testing. SE, standard error; P/T: phentermine/topiramate. Changes in HbA1c Analyses of HbA1c change included 973 treatment episodes with both baseline and 6-month HbA1c measurements (Table 7). Among episodes with available HbA1c data at baseline and 6 months, all three drugs were associated with reductions in HbA1c, particularly in patients with baseline diabetes. In multivariable linear regression, liraglutide was associated with a significantly greater decrease in HbA1c compared with orlistat (adjusted difference, −0.26 percentage points; SE, 0.07; p < 0.01). Phentermine/topiramate showed a smaller, borderline reduction in HbA1c compared with orlistat (−0.12 percentage points; SE, 0.07; p = 0.07) (Table 7). Table 7Change in glycated hemoglobin (%) from baseline to 6 months by drug group: multivariable and propensity score—weighted comparisons (results by multiple linear regression [n = 973])VariableEstimateSEp-valueIntercept (orlistat based)0.0470.1920.81Phentermine/Topiramate−0.1220.0660.07Liraglutide−0.2610.068< 0.01Age (yr)0.0020.0020.33Sex (male = 1)0.0630.0550.25BMI (kg/m2)−0.0050.0050.31DM−0.2610.054< 0.01Hypertension−0.0710.0580.23Dyslipidemia−0.0640.0520.22SE, standard error; BMI, body mass index; DM, type 2 diabetes mellitus. Change in glycated hemoglobin (%) from baseline to 6 months by drug group: multivariable and propensity score—weighted comparisons (results by multiple linear regression [n = 973]) SE, standard error; BMI, body mass index; DM, type 2 diabetes mellitus. Estimated marginal means of HbA1c change were −0.21% with orlistat, −0.33% with phentermine/topiramate, and −0.47% with liraglutide (Table 8). Pairwise multivariable comparisons showed that liraglutide reduced HbA1c more than orlistat (−0.26 ± 0.07 percentage points; p < 0.01) and more than phentermine/topiramate (0.14 ± 0.06 percentage points; p = 0.05). In propensity score—weighted analyses, however, the difference in HbA1c change between liraglutide and orlistat was attenuated and no longer statistically significant (−0.07 ± 0.08 percentage points; p = 0.36), suggesting that confounding by indication may explain part of the apparent advantage for liraglutide (Table 9). Table 8Change in glycated hemoglobin (%) from baseline to 6 months by drug group: multivariable and propensity score—weighted comparisons (estimated marginal means)DrugMean ± SE95% CIOrlistat−0.212 ± 0.055−0.319, −0.105Phentermine/Topiramate−0.334 ± 0.042−0.417, −0.251Liraglutide−0.472 ± 0.044−0.560, −0.385SE, standard error; CI, confidence interval. Change in glycated hemoglobin (%) from baseline to 6 months by drug group: multivariable and propensity score—weighted comparisons (estimated marginal means) SE, standard error; CI, confidence interval. Table 9Change in HbA1c (%) from baseline to 6 months by drug group: multivariable and propensity score–weighted comparisons (pairwise comparison)ComparisonModelDelta change of HbA1c at 6 monMean ± SE95% CIp-valueLira - OrlistatMultivariable−0.261 ± 0.068−0.420, −0.102< 0.01P/T - OrlistatMultivariable−0.123 ± 0.067−0.278, 0.0330.16Lira - P/TMultivariable−0.138 ± 0.059−0.276, −0.0010.05Lira - OrlistatPS-weighted−0.071 ± 0.078−0.223, 0.0810.36HbA1c, glycated hemoglobin; SE, standard error; CI, confidence interval; Lira, liraglutide; P/T, phentermine/topiramate; PS, propensity score. Change in HbA1c (%) from baseline to 6 months by drug group: multivariable and propensity score–weighted comparisons (pairwise comparison) HbA1c, glycated hemoglobin; SE, standard error; CI, confidence interval; Lira, liraglutide; P/T, phentermine/topiramate; PS, propensity score. Treatment persistence All 1,910 treatment episodes contributed to the analysis of treatment persistence (Table 10). Patients who did not discontinue treatment within 365 days were censored at the earlier of 365 days after the index date or the end of available follow-up. Kaplan–Meier curves showed similar persistence for phentermine/topiramate and orlistat, with a gradual decline in ongoing treatment over time, whereas liraglutide was associated with earlier and more frequent discontinuation (Fig. 3). Table 10Mean treatment time estimated as restricted mean survival time truncated at 365 days (estimated mean continuation time [days])DrugMean ± SE95% CIOrlistat143.4 ± 3.3137.1–149.9Phentermine/Topiramate143.5 ± 3.2136.9–150.0Liraglutide114.6 ± 2.8109.1–120.1SE, standard error; CI, confidence interval. Mean treatment time estimated as restricted mean survival time truncated at 365 days (estimated mean continuation time [days]) SE, standard error; CI, confidence interval. Kaplan–Meier curves for 12-month treatment persistence by anti-obesity medication. Survival curves show time to treatment discontinuation. Patients remaining on therapy at 365 days are censored. At 12 months, the probabilities of remaining on treatment were 28.8% for orlistat, 24.0% for phentermine/topiramate, and 16.6% for liraglutide. RMST analyses up to 365 days showed that orlistat and phentermine/topiramate had nearly identical mean continuation times (143.4 vs. 143.5 days), whereas liraglutide had a shorter mean continuation time (114.6 days). Liraglutide users thus spent on average 28.8 fewer days on treatment than orlistat users (95% CI, −37.3 to −20.3) and 28.9 fewer days than phentermine/topiramate users (95% CI, −37.3 to −20.5) (Table 11). Exploratory Cox models suggested a higher hazard of discontinuation with liraglutide compared with orlistat, but strong violations of the proportional hazards assumption led us to emphasize RMST and time-specific persistence estimates. Table 11Mean treatment time estimated as restricted mean survival time truncated at 365 days (pairwise comparison)ComparisonMean (days)95% CIp-valueLira - Orlistat−28.8−37.3, −20.3< 0.01P/T - Orlistat0.1−9.0, 9.20.99P/T - Liraglutide28.920.5, 37.3< 0.01CI, confidence interval; Lira, liraglutide; P/T, phentermine/topiramate. Mean treatment time estimated as restricted mean survival time truncated at 365 days (pairwise comparison) CI, confidence interval; Lira, liraglutide; P/T, phentermine/topiramate. DISCUSSION In this real-world retrospective cohort study from a large tertiary hospital in Korea, we compared three long-term anti-obesity medications (orlistat, phentermine/topiramate, and liraglutide) with respect to body weight, HbA1c, and treatment persistence. Several key findings emerged. First, all three agents produced clinically meaningful weight loss over 6 to 12 months in routine practice, but phentermine/topiramate consistently achieved the greatest magnitude of weight reduction. In both multivariable and propensity score–weighted analyses, phentermine/topiramate was associated with significantly greater mean percentage weight loss at 6 months than orlistat and higher odds of achieving ≥ 5% weight loss. Longitudinal mixed-effects models confirmed that phentermine/topiramate achieved the deepest and most sustained weight reductions at 3, 6, and 12 months. These findings are consistent with prior reviews and meta-analyses reporting that phentermine/topiramate has among the largest weight-loss effects of currently approved anti-obesity medications [4,6,7,15,16]. Second, orlistat and liraglutide produced broadly similar weight-loss trajectories in our real-world cohort, even though liraglutide users tended to have higher baseline BMI and more comorbidities. In conventional multivariable models, differences in percentage weight loss between liraglutide and orlistat were small and generally not statistically significant. Propensity score–weighted analyses suggested that liraglutide may confer a modest advantage in achieving ≥ 5% weight loss compared with orlistat, but the magnitude of this effect was much smaller than that observed for phentermine/topiramate. Taken together, these findings indicate that in daily practice, weight-loss effectiveness of liraglutide 3.0 mg is broadly comparable to that of orlistat when accounting for differences in patient characteristics. Third, with respect to glycemic control, liraglutide appeared to provide greater reductions in HbA1c than orlistat in unweighted analyses, consistent with its mechanism of action and prior evidence in patients with type 2 diabetes [8,9,10]. However, after more rigorous adjustment for baseline glycemic status and comorbidities using propensity score weighting, the difference in HbA1c change between liraglutide and orlistat was attenuated and no longer statistically significant. This suggests that at least part of the apparent glycemic benefit of liraglutide is attributable to confounding by indication, in which clinicians preferentially prescribe liraglutide to patients with poorer glycemic control or higher metabolic risk. Even so, the direction of the effect remained favorable for liraglutide, and in selected patients with diabetes or prediabetes, liraglutide may still offer clinically meaningful glycemic advantages that are not fully captured by our models. Fourth, we observed notable differences in treatment persistence. RMST and Kaplan–Meier estimates showed that phentermine/topiramate and orlistat, both oral agents, had similar 12-month persistence, whereas liraglutide had substantially lower persistence, with only about one in six episodes still on treatment at 1 year and an average of nearly one month less time on therapy. These patterns likely reflect real-world barriers to the long-term use of injectable anti-obesity medications, including injection burden, gastrointestinal adverse effects, cost, and reimbursement constraints in the Korean context [4,13,14]. Lower persistence may partially offset the potential benefits of liraglutide on glycemic control and weight, emphasizing the importance of considering both efficacy and usability when choosing an anti-obesity medication. Our findings align with and extend prior real-world and clinical trial evidence. Previous meta-analyses and real-world studies have reported robust weight-loss effects for phentermine/topiramate, moderate effects for orlistat and liraglutide, and variable effects on cardiometabolic risk factors such as blood pressure, lipids, and glycemia [7,10,15,16,17]. The present study adds to this literature by providing head-to-head comparative data across three widely used agents within a single health-care system, using detailed EHR data to assess both weight change and glycemic outcomes, as well as treatment persistence. From a clinical standpoint, our results suggest that: • Phentermine/topiramate may be preferred in patients whose primary therapeutic goal is maximal weight loss and who can tolerate a sympathomimetic and antiepileptic combination, given its superior weight-loss performance and persistence comparable to orlistat. • Phentermine/topiramate may be preferred in patients whose primary therapeutic goal is maximal weight loss and who can tolerate a sympathomimetic and antiepileptic combination, given its superior weight-loss performance and persistence comparable to orlistat. • Liraglutide may be particularly considered in patients with coexisting type 2 diabetes or high cardiometabolic risk, where potential glycemic and cardiovascular benefits are clinically important [8,9,10], but clinicians should be aware of lower treatment persistence and address barriers to adherence. • Liraglutide may be particularly considered in patients with coexisting type 2 diabetes or high cardiometabolic risk, where potential glycemic and cardiovascular benefits are clinically important [8,9,10], but clinicians should be aware of lower treatment persistence and address barriers to adherence. • Orlistat remains a viable option, especially for patients who prefer a non-systemic oral agent and may have contraindications to or concerns about the other drugs, though its weight-loss effect appears modest relative to phentermine/topiramate [6,17]. • Orlistat remains a viable option, especially for patients who prefer a non-systemic oral agent and may have contraindications to or concerns about the other drugs, though its weight-loss effect appears modest relative to phentermine/topiramate [6,17]. Ultimately, drug selection should integrate efficacy, safety, comorbidities, route of administration, cost, reimbursement policies, and patient preferences. Our real-world findings can support shared decision-making by providing more realistic expectations of weight loss and persistence for each agent beyond clinical trial settings. This study has several strengths. We used a large, well-curated EHR database from a major tertiary hospital, capturing detailed longitudinal information on prescriptions, weight, and laboratory values over several years. We assessed multiple clinically relevant outcomes, including short- and intermediate-term weight change, the attainment of clinically meaningful weight-loss thresholds, changes in HbA1c, and treatment persistence. Our statistical approach combined multivariable regression, mixed-effects models, and propensity score–weighted analyses to mitigate confounding and capture longitudinal dynamics. However, the study also has important limitations. First, as an observational retrospective study, residual confounding by unmeasured factors such as diet, physical activity, socioeconomic status, or clinician preferences cannot be excluded, even with propensity score methods [11,12,15]. Second, weight and laboratory measurements were obtained opportunistically as part of routine care rather than at protocol-mandated visits. Although we used prespecified windows around the 3-, 6-, and 12-month time points and selected the measurement closest to each nominal time, some time-window bias may remain and could slightly over- or underestimate true changes in weight and HbA1c. Third, treatment episodes were defined based on prescribing records rather than confirmed medication intake, and our definition of discontinuation may not fully capture non-adherence or intermittent use. Fourth, we could not fully ascertain or adjust for concomitant antidiabetic medications or changes in antidiabetic regimens during follow-up, which may influence HbA1c trajectories and may have differed across treatment groups; this is an important potential source of residual confounding in the glycemic analyses. Finally, some laboratory outcomes, particularly lipids and liver enzymes, were missing for a substantial proportion of episodes, limiting power and potentially introducing selection bias in those analyses, and our data derive from a single tertiary center in Korea, which may limit generalizability to other care settings. Despite these limitations, our results offer valuable insights into the comparative real-world performance of long-term anti-obesity medications in an Asian population and may inform clinical practice and policy decisions regarding obesity pharmacotherapy. In this real-world retrospective cohort study from a large tertiary hospital in Korea, phentermine/topiramate demonstrated the greatest and most consistent weight-loss effectiveness among three long-term anti-obesity medications, while orlistat and liraglutide achieved similar overall weight reduction. Liraglutide showed a tendency toward greater HbA1c reduction in higher-risk patients, although this advantage was attenuated after adjustment for confounding, and long-term persistence with liraglutide was lower than with oral agents. These findings underscore the importance of considering both weight-loss efficacy and treatment persistence, as well as individual patient characteristics and comorbidities, when selecting pharmacologic therapy for obesity in routine clinical practice. Future prospective and multi-center real-world studies, including newer agents such as semaglutide and tirzepatide, are warranted to further refine personalized obesity treatment strategies. ACKNOWLEDGMENTS We thank the Data Convergence Team (Research Information Unit) and the ABLE support staff at Asan Medical Center for their assistance with data extraction and management. Footnotes Funding: This work was conducted during the research year of Chungbuk National University in 2024. Conflict of Interest: - Authors: Nothing to declare - Reviewers: Nothing to declare - Editors: Nothing to declare Reviewer: This article was reviewed by peer experts who are not TCP editors. Author Contributions: Conceptualization: Kim CG, Koh JM, Bae KS.Data curation: Bae KS.Formal analysis: Bae KS.Investigation: Kim CG, Bae KS.Methodology: Bae KS.Project administration: Koh JM, Bae KS.Resources: Koh JM, Bae KS.Software: Bae KS.Supervision: Koh JM, Bae KS.Validation: Koh JM, Bae KS.Visualization: Bae KS.Writing - original draft: Kim CG.Writing - review & editing: Kim CG, Koh JM, Bae KS. Conceptualization: Kim CG, Koh JM, Bae KS. Conceptualization: Kim CG, Koh JM, Bae KS. Data curation: Bae KS. Data curation: Bae KS. Formal analysis: Bae KS. Formal analysis: Bae KS. Investigation: Kim CG, Bae KS. Investigation: Kim CG, Bae KS. Methodology: Bae KS. Methodology: Bae KS. Project administration: Koh JM, Bae KS. Project administration: Koh JM, Bae KS. Resources: Koh JM, Bae KS. Resources: Koh JM, Bae KS. Software: Bae KS. Software: Bae KS. Supervision: Koh JM, Bae KS. Supervision: Koh JM, Bae KS. Validation: Koh JM, Bae KS. Validation: Koh JM, Bae KS. Visualization: Bae KS. Visualization: Bae KS. Writing - original draft: Kim CG. Writing - original draft: Kim CG. Writing - review & editing: Kim CG, Koh JM, Bae KS. Writing - review & editing: Kim CG, Koh JM, Bae KS. References