Clinical outcomes associated with anti‐obesity medications in real‐world practice: A systematic literature review

Summary Anti‐obesity medications (AOMs) are efficacious and well tolerated in randomized controlled trials, but findings may not be generalizable to routine clinical practice. This systematic literature review aimed to identify real‐world (RW) evidence for AOMs to treat adults ( ≥ 18 years) with obesity or overweight (BMI  ≥  27 kg/m2). Searches conducted in MEDLINE, Embase, Health Technology Assessment (HTA) Database, National Health Service (NHS) Economic Evaluation Database, and Cochrane Central Register of Controlled Trials for studies of relevant FDA‐approved AOMs yielded 41 publications. Weight loss (WL) was consistently observed, with 14% to 58.6% of patients achieving ≥ 5% WL on orlistat, phentermine/topiramate, naltrexone/bupropion, phentermine, or liraglutide in studies of 3–6 months' duration where this was measured. When cardiometabolic risk factors were assessed, AOMs reduced or had no impact on blood pressure, lipids, or glycemia. RW data on the impact of AOMs on existing obesity‐related comorbidities and mortality were generally lacking. AOMs were associated with various adverse events, but these were of mild to moderate severity and no unexpected safety signals were reported. A pattern of poor adherence and persistence with AOMs was observed across studies. Overall, the review confirmed the effectiveness of AOMs in RW settings but demonstrated large gaps in the evidence base.


| INTRODUCTION
Obesity is a major public health issue with a prevalence that has tripled over the last 45 years. 1 In 2015, it was estimated that nearly 604 million adults (12%) worldwide were classified as having obesity (body mass index [BMI] ≥ 30 kg/m 2 ). 2 Furthermore, in an analysis of the 2015 Global Burden of Disease study, high BMI was reported to account for 4 million deaths globally and to contribute to 120 million disabilityadjusted life years. 2 Obesity also imposes a considerable economic burden on healthcare systems and society, 3 primarily driven by the treatment of obesity-related chronic diseases as well as presenteeism, absenteeism, and reduced employment rates. 4 For example, in the United States, individuals with obesity had annual healthcare costs US $3500 higher than individuals without obesity, resulting in a national cost of US$316 billion per year or 27.5% of US healthcare spending in 2010. 5 Similarly, international data from 52 Organisation for Economic Co-operation and Development (OECD) countries suggests that over the next 30 years, overweight and obesity will cost US$425 billion per year, representing 8.4% of total global healthcare spending. 4 Prevention of obesity through policy changes and healthy lifestyle promotion is critical to curb the worsening epidemic. However, with such high proportions of individuals already manifesting obesity, there is also a pressing need for treatment. A stepwise approach to obesity treatment is generally advocated involving initial lifestyle interventions followed by pharmacologic intervention and bariatric surgery, if necessary. Lifestyle-based therapies represent the cornerstone of obesity management, but alone do not provide sustainable weight loss in most individuals, 6 and bariatric surgery, though highly effective, is applied in only a minority of eligible cases. 7 8,9 Furthermore, another four treatments (phentermine, benzphetamine, diethylpropion, and phendimetrazine) are FDA approved for short-term (a few weeks) use, although with the exception of phentermine these are rarely utilized in real-world settings. 10 The efficacy and safety of AOMs have been well documented in randomized controlled trials (RCTs). A systematic literature review including 35 RCTs reported that the AOMs FDA-approved for longterm use at the time were all associated with greater weight loss and weight-loss maintenance compared with placebo and were associated with generally low rates of serious adverse events (SAEs). 11 However, the effectiveness of AOMs in real-world practice is not as well understood. Unlike RCTs, real-world studies include heterogeneous patient samples that are more representative of the general disease population likely to be treated by primary care and specialist physicians.
Real-world studies can support data from RCTs and provide more information on clinical outcomes, safety signals, patient persistence and adherence, economic outcomes, and longer-term treatment trends, all of which are fundamental in informing disease management practices and healthcare policy. 12 The objective of the current review was, therefore, to identify, summarize, and interpret retrospective or prospective published studies that provide real-world evidence (RWE) for AOMs in the treatment of adults ( ≥ 18 years) with obesity or overweight. While the original search comprised a broad focus, this manuscript is limited to a summary of weight change, cardiometabolic risk factors, adverse events (AEs), and adherence, persistence, and discontinuation, since these were the most commonly and consistently reported measures.

| METHODS
A robust and reproducible protocol for the literature search was developed that detailed the proposed approach, objectives, search strategy, study selection criteria, methods for data extraction and synthesis, and outcomes of interest that were specified a priori. The protocol reduced the potential impact of review author bias, ensured transparency and accountability, and maximized the chances of accurate data extraction. The overall search strategy comprised three concepts: "weight loss" AND "specific AOMs of interest" AND "RWE." Notably, the more general concept of "AOMs" without mention of specific drugs of interest was not a part of the search strategy as the aim was to only identify and include studies in which drug-level data for the specific AOMs of interest were presented. Concepts were captured using subject headings and text-word searches in the title, abstract, and keyword-heading fields. A base-case strategy was developed for MEDLINE and adapted to the other databases (Tables S1-S5); additional details regarding the search strategy can be found in the Supporting Information.

| Eligibility criteria
The search eligibility criteria are shown in Table 1. While the original search included a range of AOMs, only those that were FDA-approved for long-term use at the time of the search are the focus of the current article. Publications that evaluated outcomes associated with lorcaserin and sibutramine are not summarized here, but where evaluated as comparators in the included studies, findings were noted. Of the AOMs FDA-approved for short-term use, only phentermine was included as it is one of the most frequently prescribed in real-word practice. 10

| Study selection process
Search results were assessed independently by two reviewers, using a two-phase approach that consisted of (1) a broad review of the title and/or abstract of search results and (2) a subsequent full-text review of potentially eligible studies identified at Stage 1. Any studies failing to meet the selection criteria at Stage 2 were excluded and the reason for exclusion recorded. Any disagreements between reviewers were resolved by discussion until consensus was met.
Data extraction was performed on a standardized data extraction form by two reviewers, with quality checking by a third. Variables extracted included study population, interventions, study type and methods (including data source), study duration, and specific outcomes data.

| RESULTS
The search identified 2613 studies for eligibility review after removal of duplicates, of which 2535 were excluded following review of titles and abstracts. Of 78 full-text records, 35 were excluded ( Figure 1). An additional two studies were identified by citation searching of included records to yield a total of 45 studies. Four of these studies evaluated sibutramine alone and so were also excluded, leaving 41 eligible studies for inclusion in the review. Table 2 provides an overview of the characteristics of included studies. Studies were conducted across a wide geography, with the United States and the United Kingdom being the most represented countries.

| Study designs
Most studies were of a retrospective design (n = 33), and data were mostly collected from medical records and charts (electronic or otherwise). Other data sources utilized in retrospective studies included administrative claims databases, pharmacy prescription data, and AE reporting systems. Only eight studies were prospective in design, with data mostly obtained at prespecified clinic visits. Studies were conducted specifically in primary (n = 12) and secondary/tertiary care settings such as specialist clinics, academic centers, and hospitals (n = 16), with 13 studies including data from both settings. In most studies, outcomes associated with AOMs were compared with base- Sample size for all patients who received AOMs after surgery; number of patients receiving phentermine specifically was not reported.

| Weight outcomes
Across the 28 studies including weight outcomes, regardless of study population, the most consistently reported measures were absolute weight reduction (in kg or lb; n = 22 studies), percentage reduction in body weight (n = 19), and categorical weight loss according to clinically meaningful thresholds (n = 15). The main findings from studies reporting on these measures are summarized in Table 3. Other weight outcomes included changes in BMI and waist-to-hip ratio, but these were seldom reported. Weight outcomes associated with orlistat (12 studies) or phentermine (10 studies) were the most reported; five studies reported on weight outcomes with PHEN/TPM and five with liraglutide, while only a single study included NTX/BPN (Table 3). Findings were pooled from multiple AOMs in two additional studies. 35,50

| General obesity population
Across all studies in a general obesity population, AOMs were associated with a reduction in weight regardless of study design and duration. However, the magnitude of weight loss varied considerably from study to study (Table 3). With respect to achieving a clinically meaningful weight loss, the range of patients who lost ≥ 5% of their total body weight was 22.2% in a 12-week orlistat study, 23 up to 50% in a 12-week phentermine study, 26 and > 50% in two liraglutide studies (Table 3). 32,48 Few studies were identified that directly compared different AOMs of interest. Findings from these comparative analyses are summarized in Table 4. Orlistat was associated with a significantly poorer weight-loss response compared with liraglutide after 3-6 months (p < 0.0001) 32 and a numerically lower absolute weight reduction compared with phentermine and PHEN/TPM at ≥ 20 weeks in a general obesity population. 33 Patients receiving phentermine or PHEN/ TPM were more than 50% more likely to experience a ≥ 5% weight loss compared with those receiving orlistat (p < 0.01). 33 In addition, in a matched cohort study, both phentermine and PHEN/TPM were associated with a greater weight reduction compared with NTX/BPN. 47

| Diabetes population
Among the five studies that evaluated weight response among patients with obesity and T2DM, treatment with orlistat, phentermine, PHEN/TPM, and NTX/BPN was associated with a reduction in weight that appeared to be comparable with losses observed in a general population of individuals with obesity (Table 3). 16,24,30,34,47 One study demonstrated that in patients with T2DM, orlistat in combination with participation in a clinical weight-loss program resulted in a numerically better weight-loss response compared with orlistat alone, although the difference failed to reach statistical significance (Table 3). 34 In another study, no differences were reported in weight loss between patients with or without T2DM treated with phentermine, PHEN/TPM, or liraglutide, but a difference was demonstrated with NTX/BPN (T2DM, À4.8 kg vs. non-T2DM, À2.2 kg; p = 0.05; Table 3) although patient numbers were low (n = 32). 47

| Postsurgical population
In patients with obesity who had previously undergone bariatric surgery and experienced subsequent weight regain or insufficient weight loss postoperatively, treatment with phentermine, PHEN/TPM, and liraglutide all resulted in weight reduction (Table 3). 19 (Table 3). 20 Similarly, in one study that pooled data from multiple AOMs, weight loss was greater in patients who received drug treatment following RYGB compared with those treated post-SG. 50 Very few comparative data were available in surgical patients, but one study provided evidence that phentermine may produce superior weight loss compared with PHEN/TPM in surgical patients, although it should be noted that the number of patients receiving PHEN/TPM in this analysis was small (n = 6) ( Table 4). 52

| Cardiometabolic risk factors
Cardiometabolic risk factors were less well studied among the included articles. Thirteen studies overall (around 30% of those included; general obesity population [with or without DM], n = 12; surgical population, n = 1) evaluated the impact of AOMs on parameters including blood pressure, heart rate (HR), lipids, fasting blood glucose, and glycated hemoglobin (HbA 1c ). The effects of orlistat and phentermine were evaluated most frequently, followed by liraglutide and PHEN/TPM. Results varied across studies, with some demonstrating a positive impact on cardiometabolic risk factors and others showing no effect. An overview of the trends across studies is shown in Table 5.

| Orlistat
In the six studies that evaluated the effect of orlistat on blood pressure, systolic blood pressure (SBP) was significantly reduced in three studies and numerically reduced or unchanged in three, while no effect or a numerical decrease in diastolic blood pressure (DBP) was observed in four studies and a significant reduction reported in two (Table 5). Triglycerides (TG), total cholesterol (TC), and low-density lipoprotein cholesterol (LDL-C) were generally reduced in association with orlistat treatment, while impact on high-density lipoprotein cholesterol (HDL-C) was more variable (Table 5). Glycemic parameters were consistently reduced in patients with obesity and DM (mostly T2DM) who received orlistat. 16,18,24,32,34 One German postmarketing study also evaluated the effects of orlistat on cardiometabolic risk factors in subgroups of patients with comorbidities and demonstrated that improvements in blood pressure or lipid parameters were greater in individuals with hypertension or dyslipidemia, respectively. 18 In a single comparative study, no clinically significant differences from baseline to 6 months in blood pressure, lipids, or HbA 1c was observed between patients treated with orlistat, phentermine, or PHEN/TPM. 33

| Phentermine and PHEN/TPM
Like orlistat, phentermine appeared to be associated with a reduction in SBP (Table 5). However, small increases in HR from baseline were reported in phentermine-treated patients, although this did not reach statistical significance. 36,40 Two studies reported on the effect of PHEN/TPM on cardiometabolic risk factors, with few changes in blood pressure, lipids, or glycemia observed (Table 5). 33,42

| Liraglutide
Few data (n = 2 studies) were identified regarding the impact of liraglutide on cardiometabolic risk factors (Table 5). Where studied, liraglutide was generally associated with a reduction in blood pressure, lipids, and glycemic parameters. 32,48

| Postsurgical patients
No changes in lipid or glycemic parameters were reported in a single study including patients who received phentermine for weight gain/ insufficient weight loss after bariatric surgery (Table 5). 49

| Existing comorbidities
Five of the identified studies in a general obesity population (with or without T2DM) also evaluated the impact of AOMs on existing comorbidities, which was generally reported as a change in specific medications. For example, antihypertensive, glucose-lowering, and lipid-lowering drug use was reported to be reduced following orlistat initiation in three studies including patients with obesity and comorbid diseases, 16 identified for NTX/BPN. Where reported, AEs appeared to be mild to moderate in severity and were mostly short-lived.

| Orlistat
In orlistat studies, AEs affecting the gastrointestinal system were the most commonly reported events. 17

| Liraglutide
AEs associated with liraglutide were reported in two studies in a general obesity population. 19,32 The most common AEs in liraglutidetreated patients were mostly gastrointestinal in nature, including nausea and vomiting, and diarrhea.

| Postsurgical patients
Two studies evaluated AEs associated with liraglutide in patients who had previously undergone bariatric surgery. Among the most commonly reported AEs were nausea, headache, constipation, and diarrhea. 20,51 3.6 | Adherence, persistence, and discontinuation  Table 6. Across studies, adherence, persistence, and discontinuation were measured in multiple different ways, rendering it impossible to compare outcomes from one investigation to the other. However, there was a general pattern of poor compliance with all AOMs. For example, in a US retrospective observational cohort study using data from the Veterans Affairs Corporate Data Warehouse that used the medication possession ratio (MPR) to determine 6-month adherence, the highest rate reported was only 38.2% in PHEN/TPM-treated patients, with other AOMs performing even more poorly (  (Table 6). [31][32][33]35 In one study, more patients remained on liraglutide at 12 months versus orlistat (p = 0.011) and at the end of follow-up persistence was higher (55% vs. 19.5%; p < 0.0001). 23 However, after adjustment for baseline factors, there was no significant difference between the persistence curves. A significantly lower risk of discontinuation with liraglutide was demon- Some of the data from patient subpopulations warrants further discussion. In contrast to RCTs, which have consistently shown less weight loss with AOMs in populations with T2DM compared to those without T2DM, the few real-world studies that evaluated these drugs  31 Both providers and patients also tend to view AOMs as a jump start for weight loss rather than chronic therapy that extends to weight maintenance, and this may account for lack of persistence even in those who initially achieve meaningful weight loss. Since the benefits of short-term weight loss are unclear, low compliance with AOMs raises important questions regarding the cost-effectiveness or value of the treatment as it is currently applied in the real world. Low persistence and adherence will need to be addressed to sustain the observed real-world effectiveness of AOMs and achieve the potential long-term benefits of AOM-induced weight loss.
Real-world data are emerging as an important component of the overall evidence base for understanding the utility of medications across a range of patient populations. 12 These data may represent a valuable supplement to those obtained in RCTs. For example, sibutramine was withdrawn from global markets due to CV safety concerns reported in an RCT. 55 Real-world studies failed to demonstrate such CV risks in a more generalizable patient population, suggesting that the marketing authorization for sibutramine may have been inappropriately withdrawn for patients without pre-existing CV disease. 13,56,57 More studies identified by the current search were conducted in secondary/tertiary care compared with primary care settings. However, as obesity rates continue to climb and its acknowledgment as a chronic disease continues to grow, more and more individuals will seek weight management advice from their primary care physician.
Therefore, it is important to gain a better understanding of the experience of patients in this setting. The fact that 12 primary care studies were identified by this review suggests that AOMs are effective and well tolerated in this setting. These studies provide valuable information regarding the translation of obesity management from the specialist to the generalist setting and the feasibility of scaling the pharmacologic management of obesity.
One of the challenges with RWE is the difficulty in interpreting data across studies. Methods, populations, data collection, and reporting vary considerably from one evaluation to the next. In addition, con- with T2DM because few of these provided details of any concomitant glucose-lowering medications. Since many of these agents also promote weight gain or weight loss, they could have an impact on AOM effectiveness in these patients that confounds the results.
The current review is subject to several limitations that relate to the search itself, the evidence base, and issues inherent in the methods of real-world studies. While the search was conducted using a robust and reproducible protocol, the approach was largely pragmatic, and it cannot be ruled out that other studies relevant to the research question may have been published. The research question focused on specific FDA-approved AOMs deemed to be relevant to the current pharmacologic management of obesity. As such, studies that provided RWE for the effectiveness of AOMs generally without specific drug-level data were not a part of the search strategy. In addition, a two-stage approach was adopted for the review of search results; at the first stage, the decision to include or exclude a publication is made based on review of the title/abstract and not on a comprehensive review of the full-text of the article, so it is possible that potentially relevant studies are excluded at this stage due to lack of detail in the title or abstract. For example, if the AOMs of interest were not specifically mentioned by name in the title/abstract, the study did not meet our eligibility criteria, but it could be that the fulltext of the publication did provide disaggregated data on that agent.
Furthermore, inconsistencies in the description of RWE in the literature, the range of terminologies used, and the lack of clarity in methods for data collection-even in the full-text of some papersmade the decision to include challenging in some cases. The reviewers were also compelled to exercise a level of value judgement as to whether a study truly reflected real-world practice. For example, some studies-though conducted in a clinical setting and termed observational-had strict inclusion/exclusion criteria, highly prescriptive scheduling and conduct of clinic visits, and did not appear to fully reflect patient behaviors were they not to have been included in the study. In such cases, the studies were extensively discussed among reviewers until consensus was met.
While the search was designed to identify a wide array of outcomes associated with AOMs, substantial gaps were evident in the RWE. Few studies were identified that reported data on outcomes other than weight change, AEs, and compliance. Data on cardiometabolic risk factors was generally limited, although this is perhaps not surprising given that healthcare providers may not routinely monitor metabolic labs in clinical practice due to the constraints of cost and insurance coverage. Only two studies each (5% of the total) were identified as including limited data on economic outcomes 16,49 or patient-reported outcomes. 17,18 This is also expected since routine real-world data sources like EMR or claims databases will often not capture patient-reported outcomes. There is, therefore, a need for other forms of data capture in a real-world setting (e.g., in the form of pragmatic trials) to evaluate these types of outcomes.
Another identified gap was that most studies were single arm, comparing the impact of each drug to baseline; practically none included a control arm, and relatively few studies were identified that directly compared different AOMs. Furthermore, few studies included details regarding the analytical approach for handling missing data.
Since these methods can influence bias in the results, it is difficult to know if accurate conclusions about the data have been drawn in individual studies. Many of the included studies also contained small numbers of patients and were of short duration; data are, therefore, lacking on the maintenance of weight loss. Finally, selection bias is inherent in many of the studies since healthcare professionals often do not proactively address obesity with the use of AOMs, but rather patients seeking weight-loss options beyond diet and exercise request medication from their physician or self-refer to weight-management specialists. It is not known from the included studies whether physicians provided objective advice and counseling regarding AOMs when medically indicated independent of patient requests. As such, the data may not represent the real-world effectiveness of AOMs in the indicated population, but rather in a subset of patients who may have proactively requested this treatment option.

| CONCLUSIONS
RWE for the effectiveness and safety of AOMs were identified in a diverse obesity population. Such evidence can supplement the findings previously reported in tightly controlled RCT patient samples.
Across studies employing prospective and retrospective designs, AOMs were consistently demonstrated to reduce weight in a general population of patients with obesity/overweight, in patients who had regained weight or experienced inadequate weight loss after bariatric surgery, and in specific patient subgroups such as T2DM. Weight loss was often accompanied by positive changes in other cardiometabolic risk factors, when measured. Although AOMs were well tolerated in real-world studies with mostly mild to moderate AEs, a general pattern of poor compliance was apparent with all treatments, the reasons for which will need to be better understood and addressed to fully evaluate the long-term benefit of AOMs in the real world.
Importantly, the review identified large gaps in the evidence base for AOMs in treating patients with obesity or overweight in real-world practice, including few comparative effectiveness studies and a narrow range of reported outcomes. Real-world studies are also affected by the same issues that plague RCTs in the obesity field with respect to untangling the interactions between adjunct lifestyle measures and AOMs. There is a clear need for more extensive and consistently designed real-world studies, including pragmatic trials, that incorporate valid control and/or comparator groups, that examine more recently approved medications, and that more robustly account for the relative contributions of lifestyle interventions. Such studies can capture a broader range of outcomes, including cardiometabolic, economic, and patient-reported measures. Strengthening the approach to RWE generation in obesity will help build a more accurate picture of the value of AOMs in routine clinical practice, especially as newer agents promising greater efficacy are on the horizon.