Examining paradoxical session attendance and weight loss relationships in a clinic based lifestyle modification intervention

Abstract Objective Evaluations of lifestyle modification interventions (LMIs), modeled after the Diabetes Prevention Program, have repeatedly shown a dose‐response relationship between session attendance and weight loss. Despite this, not all participants had “average” weight loss experiences. Nearly one‐third of LMI participants experienced unexpected, paradoxical outcomes (i.e., high attendance with little weight loss, and low attendance with clinically significant weight loss). Paradoxical weight‐loss outcomes were characterized based on session attendance among participants in a group‐based LMI in a real‐world healthcare setting. This group‐based LMI was delivered over 1 year to participants with the possibility of attending up to 25 sessions total. Methods LMI participants identified in 2010–2017 from electronic health records were characterized as having low (<75%) or high (≥75%) session attendance. Weight‐loss outcomes were defined as expected (≥5%, high‐attendance; <5%, low‐attendance) or paradoxical (≥5%, low‐attendance; <5%, high‐attendance). Paradoxical‐outcome‐associated characteristics were identified using logistic regression. Results Among 1813 LMI participants, 1498 (82.6%) had low and 315 (17.4%) high session attendance; 555 (30.6%) had paradoxical outcomes, comprising 415 (74.8%) responders (≥5% weight‐loss) and 140 (25.2%) non‐responders (<5% weight‐loss). Among participants with high session attendance, paradoxical non‐responders were more likely to be female (odds ratio [OR]: 2.76; 95% confidence interval [CI]: 1.32, 5.77) and have type 2 diabetes (OR: 3.32; 95% CI: 1.01, 10.95). Among low‐attendance participants, paradoxical responders were more likely to be non‐Hispanic White and less likely to be non‐Hispanic Black (OR: 0.35; 95% CI: 0.18, 0.69), non‐Hispanic Asian (OR: 0.40; 95% CI: 0.22, 0.73), or Hispanic (OR: 0.53; 95% CI: 0.35, 0.80). Conclusions In a healthcare setting, nearly one‐third of LMI participants experienced paradoxical outcomes. More research is needed to understand the facilitators and barriers to weight loss above and beyond session attendance.


| INTRODUCTION
In 2010, the U.S. Congress authorized the Centers for Disease Control (CDC) to enact the National Diabetes Prevention Program (NDPP), a collaboration of private and public organizations from across the nation promoting the dissemination of lifestyle modification interventions (LMIs) to achieve moderate sustainable weight loss, reduce the risk of developing type 2 diabetes, and improve overall health. 1 CDC-aligned LMIs were modeled after the efficacious Diabetes Prevention Program (DPP), 2 an intervention tested via a randomized controlled trial (RCT) of an intensive individually administered LMI focused on the reduction of overall calorie and fat intake as well as adoption of moderate regular physical activity.The DPP trial resulted in an average weight loss of 7% and decreased the risk of onset of type 2 diabetes by 58% during 2.8 years of follow-up in adults with impaired glucose tolerance compared with a placebo control group. 3In contrast, metformin decreased the risk of type 2 diabetes by 31% and was found to be significantly inferior to the LMI. 4 Studies of CDC-aligned translational interventions, [5][6][7][8][9] which deliver year-long group-based LMIs of 16 sessions in the first 6 months followed by roughly once-per-month maintenance sessions, have shown a dose-response relationship between session attendance and weight loss.Retention in these programs is a persistent issue; for example, Ely and colleagues 6 found that among groupbased NDPP LMI participants across multiple cohorts of more than 14,000 middle-aged adults, only 43% completed all 16 sessions of the program compared to 95% from the original DPP trial. 10Nevertheless, session attendance was positively associated with weight loss, with each NDPP session attended corresponding on average to a 0.31% reduction in body weight (or approximately 5% loss of body weight over the course of 16 sessions). 6aluations from LMIs, modeled after the DPP, have repeatedly shown a dose-response relationship between session attendance and weight loss.Despite a dose-response relationship at the population level, weight loss is a heterogeneous phenomenon at the subgroup or the individual level where not all participants have "average" weight loss experiences.Some individuals experience unexpected, seemingly paradoxical outcomes (i.e., high attendance with little weight loss, and low attendance with clinically significant weight loss).2][13] For example, our prior work showed that on average, group-based LMI participants lost 3.7% (95% CI: −3.9%, −3.5%) body weight at 12-week follow-up, but clustering analyses revealed three distinct patterns of weight loss. 14ese were characterized as minimal-to-no weight loss, delayedminimal weight loss, and steady moderate weight loss, equivalent to mean weight changes of 0.4%, −2.3%, and −4.8% at a 12-week follow-up, respectively. 14Notably, these short-term changes in weight were highly correlated with weight loss outcomes at the completion of the program (52-week follow-up) and were dependent upon levels of participation.However, despite the average direct relationship between weight loss and session attendance, not all participants had "average" weight loss experiences.In fact, some with low attendance achieved program-defined weight loss goals, whereas others with high attendance did not.Understanding these unexpected and paradoxical outcomes is important, especially for participants who attend most (or all) sessions but do not achieve clinically meaningful weight loss (defined as 5% of body weight). 15To our knowledge, there are no studies that have examined these paradoxical outcomes.
Using data from electronic health records (EHR) of a large integrated healthcare system in Northern California, the authors sought to examine session attendance and weight-loss outcomes of participants in a CDC-aligned group-based LMI to identify and characterize patterns of paradoxical outcomes.The hypothesis was that such outcomes could be associated with differences in comorbidity burden and overall metabolic risk at baseline.

| METHODS
A longitudinal descriptive analysis of LMI-enrolled participants 9 using EHR data from an integrated healthcare delivery system serving a geographically diverse patient population in northern California, United States (2010-2017) was conducted.Details of the lifestyle modification program are described elsewhere (Huang et al. 16 ).
Patients were categorized into four mutually exclusive groups based on session attendance and weight loss goal achievement.Specifically, participants were characterized as having low (<75% of sessions completed) or high (≥75% sessions completed) attendance, and as LMI responders (≥5% weight loss) or non-responders (<5%) from baseline to 12 months follow-up.The 75% cut-point used in our analysis was derived from the assertion that 19 out of 25 sessions are considered a "good" attendance or proxy of engaged program participation.This cut-point has been used in other studies. 17This study was approved by the Sutter Health Institutional Review Board.

| Covariates
Baseline demographics, clinical characteristics, and healthcare utilization were extracted from the EHR in the 12 months prior to the index date.Demographic variables included age, primary insurance, self-reported gender, race/ethnicity, and preferred spoken language.
Race and ethnicity were defined by Hispanic ethnicity, followed by racial group.If a patient did not self-identify as Hispanic, they were classified based on their self-reported race.Throughout this article, references to racial groups imply non-Hispanic.This paper used the designation "Hispanic" instead of the commonly used term "Latinx" for consistency with US Census categories 18 and self-reported data collection.Based on program participants' available home addresses, median income by zip code was used as a proxy for socioeconomic status.Baseline clinical characteristics including weight, body mass index (BMI), smoking status, and active prescriptions.Co-morbidities were identified based on a combination of ICD 9/10 codes, medication orders, and/or laboratory values in EHR, including type 2 diabetes, prediabetes/increased risk of type 2 diabetes, metabolic syndrome, hypertension, dyslipidemia, atherosclerotic cardiovascular disease, depression, and anxiety.The Charlson co-morbidity index (CCI) was also calculated. 19The CCI score 20 has been validated and reported to have a strong ability to predict mortality. 21The higher the score, the higher likely the severity of comorbidity.The CCI score was calculated using the diagnostic code from the problem list, encounters, and billing data.Healthcare utilization included a number of outpatient visits and telephone/electronic visits, preventive visits, and documented influenza immunization.

| Outcomes
The outcomes of interest were the occurrence of a paradoxical (i.e., unexpected) outcome (Yes/No) defined as paradoxical responder: low session attendance (<75% sessions) and significant weight loss (≥5%); or paradoxical non-responder: high session attendance (≥75% sessions) and insignificant weight loss (<5%). 22Conversely, non-paradoxical (i.e., expected) outcomes were defined as expected responder: high session attendance and significant weight loss; or expected non-responders: low session attendance with low weight loss.

| Statistical analyses
The authors summarized quasi-continuous variables with means and standard deviations (SD) and categorical variables with counts and proportions for the four mutually exclusive groups: ( 1 Adjusted odds ratios (OR) and 95% confidence intervals (CI) were calculated from both logistic regression models.All statistical analysis was conducted in SAS v9.4 (SAS Institute).

| RESULTS
A total of 1813 participants met the eligibility criteria.For participants, 94.2% had a baseline weight that was measured on the date of the first LMI visit.Furthermore, 99.7% of participants had a baseline weight measured within 3 months of the initial LMI visit.Overall, 590 (32.5%) participants attained ≥5% weight loss.The majority of participants (82.6%, 1498) had low session attendance.A total of 1083 participants (86.0%) were classified as expected non-responders (low session attendance, <5% weight loss) and 175 (14.0%) as expected responders (high session attendance, ≥5% weight loss).The remaining 555 (30.6%) participants had unexpected outcomes; 415 (74.8%) were classified as paradoxical responders (low session attendance, ≥5% weight loss) and 140 (25.2%)were classified as paradoxical nonresponders (high session attendance, <5% weight loss). -64 Demographic, clinical, and healthcare utilization characteristics of participants with expected and paradoxical (unexpected) outcomes are shown in Table 1.Among participants with low session attendance, paradoxical responders (i.e., those who unexpectedly lost weight despite lower attendance) were slightly older (mean age 54.4 vs. 52.5 years, p = 0.01).The racial/ethnic composition of these participants differed; those self-identifying as non-Hispanic White were twice as likely to experience a paradoxical response as compared to non-Hispanic Black or non-Hispanic Asian participants (32% vs. 16%, p < 0.001).Paradoxical responders also tended to have a higher CCI score than expected non-responders, with 8.9% versus 5.4% scoring a CCI of 3 or more, respectively (p = 0.04).
Among those with low attendance, after adjusting for other covariates (Table 2, Model 1), a CCI score of 3 or more was associated with nearly twice the odds of being a paradoxical responder versus an expected non-responder (OR: 1.99; 95% CI: 1.13, 3.52).Similar to the unadjusted analysis, paradoxical responders were also more likely to be non-Hispanic White and less likely to be non-Hispanic Black (OR: 0.35; 95% CI: 0.18, 0.69), non-Hispanic Asian (OR: 0.40; 95% CI: 0.22, 0.73), or Hispanic (OR: 0.53; 95% CI: 0.35, 0.80).Among participants with high attendance, females had 2.76 (95% CI: 1.32, 5.77) times the odds of being a paradoxical nonresponder than an expected responder (Table 2, Model 2); those with diabetes had three times the odds of not responding as compared to those with no metabolic conditions (OR: 3.33; 95% CI: 1.01, 10.9).

| DISCUSSION
In this study, the authors sought to examine paradoxical or unexpected outcomes from a CDC-aligned group-based LMI in a large integrated healthcare system in Northern California.Given the concerns and efforts to address low engagement and long-term retention for these types of LMIs, it is worthwhile to characterize paradoxical responders and non-responders. 8,23Strikingly, it was found that only 17.4% of participants attended more than 75% of the sessions offered.Yet, 32.5% achieved ≥5% weight loss and sustained it at 12 months from baseline. 9aracteristics associated with both paradoxical response (i.e., low session attendance and ≥5% body weight loss) and non-response (i.e., high session attendance and <5% body weight loss) were explored.Paradoxical outcomes occurred for nearly one-third of the cohort (30.6%), of which 415 (74.8%) were responders (with low session attendance) and 140 (25.2%)were non-responders (with high session attendance).The rates of paradoxical outcomes within each attendance group differed; 27.7% of those in the low-attendance cohort experienced paradoxical outcomes, as compared to 44.4% of those in the high-attendance cohort.These findings underscore that the "dose-response" relationship between session attendance and weight loss is an average effect and that not all participants experience "average" outcomes; rather, weight loss is a heterogenous phenomenon.
The authors initially hypothesized that participants with high session attendance and low weight loss were more likely to have comorbid conditions and higher overall metabolic risk at baseline.These patients with type 2 diabetes were found to be three times more likely to be non-responders than those without metabolic conditions.
The LMI was designed for patients with pre-diabetes among those with higher glucose levels, and therefore may not have provided sufficient support for those already diagnosed with type 2 diabetes.Selfmanagement, including diabetes self-management education, may need to be more multifaceted for those with the greatest cardiometabolic burden.
Females within the high-attendance cohort had nearly 3-fold odds of non-response compared with males, after adjusting for other sociodemographic, clinical, and healthcare utilization factors.
Several studies 12,24,25 have found that males lose more weight than females in weight loss interventions.It is beyond the scope of this study whether paradoxical non-responders had lower levels of selfefficacy or other social drivers of health and other lifestyle factors (e.g., occupation, family support structure, sleep habits, caregiver status) that can influence health behaviors.Furthermore, there could be undiagnosed hormonal imbalances, especially among women, or mental health factors.For example, hypothyroidism, the second most common endocrine disorder among women, has been found to be associated with weight gain and obesity. 26Depression and anxiety have been associated with both weight loss [27][28][29] and weight gain 27,28,30,31 , representing a paradox that has been much studied.In our study, diagnoses of depression or anxiety were not found to be associated with paradoxical outcomes (either paradoxical response or non-response).
It is possible that genetic factors could contribute to paradoxical non-response despite high session attendance.Genetic predisposition to human obesity risk is conditioned by thousands of DNA variants that make obesity prevention and management a major challenge. 324][35][36] A study conducted by Frayling et al. (2007) showed that the intron 1, a common variant of the fat mass and obesity (FTO) gene, is associated with BMI   The baseline healthcare utilization variables were in 1 year prior to the index date.These visits were non-LMI visits.Preventive visits were defined by the CPT codes '9938', '9939', 'G0344', 'G0402', 'G0438', and 'G0439' using billing data.Influenza immunization was defined by a procedure code from the billing data.The number of outpatient visits and number of telephone/electronic visits were defined from patient encounter data.and predisposes individuals to obesity. 37In addition, Scuteri et al.

p-Value
(2007), who performed a genome association scan to identify genetic variants associated with obesity, found that FTO genetic variants are associated with increased changes in the individuals' BMI, hip circumference, and body weight. 38These changes could have a significant effect on the risk of developing obesity.These contributors underscore an urgent need for the identification of better therapeutic targets and the development of effective medications.[46][47] It is important to note that the LMI examined was only offered in-person and without a telehealth option to facilitate virtual attendance and/or remote participation.It is unclear how the acceleration of telehealth may impact engagement in LMIs for patients.Some evidence suggests that the effectiveness was lower among online diabetes prevention LMI participants compared to in-person participants. 48Furthermore, more adults referred to online programs enrolled, but fewer engaged at 6-month compared to in-person participants. 48This may, in part, be due to a need to design programs intended for telehealth delivery as opposed to adapting a program that was designed for an in-person setting.Additionally, it may be important to identify which patient segments and subgroups may benefit from this modality, given barriers as well as preferences.
More work is needed to optimize this modality for maximum effectiveness.
This study has limitations that should be considered when interpreting the findings.First, study data were derived from an open healthcare system, meaning that individuals can receive care outside the system.Accordingly, some diagnoses or other covariates may have been missed; however, the program is restricted to patients who receive primary care at Sutter Health, and therefore are considered likely to receive most of their ambulatory care within the system.
Second, because this study was retrospective and patient-reported data were not collected, behavioral factors or other social factors that may impact program engagement could not be measured.
Therefore, unmeasured confounding is likely.Despite these limitations, data were from a large health healthcare system and therefore our results have good generalizability to other health systems across the US.
LMI participants were identified as having at least one encounter for the program recorded in the EHR between 2010 and 2017.The first LMI encounter was considered the index date.Eligible cohort patients were required to be at least 18 years of age (as of the index date) and to have EHR activity during the 12-36 months prior to the index date to confirm previous contact with the system and to gather medical history.Patients were also required to have a baseline weight measurement captured on or up to 12 months prior to the index date and a follow-up weight measurement captured at 12 months (�3 months) from the index date.All participants included in the analysis enrolled in the full 12-month program with the possibility of attending up to 25 sessions total over the year.Exclusion criteria included patients diagnosed with International Classification of Disease (ICD) 9 or 10 indicating potential dramatic weight change due to pregnancy, metastatic cancer, gastric bypass surgery, chronic liver disease, or end-stage kidney disease in the 12 months prior or up to 24 months after index date (tab.A1 in Romanelli et al. 14 ).
) paradoxical responders; (2) paradoxical non-responders; (3) expected responders; and (4) expected non-responders.Comparisons were made between paradoxical response versus expected response for low session attendance and paradoxical non-response versus expected response for high session attendance, respectively.Chi-square tests and Fisher's exact tests were used for categorical variables and ttests for quasi-continuous variables.Within the low-attendance (Model 1) and high-attendance (Model 2) groups, logistic regression was used to assess the probability of paradoxical versus an expected weight-loss outcome adjusting for patient demographics and clinical characteristics.
AZAR ET AL.

1
Demographics and clinical characteristics by session attendance and expected versus paradoxical outcomes.

T A B L E 2
Odds of paradoxical outcomes from logistic model adjusted by patient demographics and clinical characteristics.

5 |
CONCLUSIONApproximately one third of participants from a group-based LMI in Northern California experienced paradoxical outcomes.Nearly half the individuals who attended at least 75% of program sessions did not achieve clinically significant weight loss (paradoxical non-responders) and were more likely to be female and have type 2 diabetes after adjusting for other factors.In contrast, those with low attendance who achieved weight-loss goal achievement (paradoxical responders) were more likely to self-report as non-Hispanic White and have higher CCI scores than those who did not.This study emphasizes that there are participants in LMIs whose weight loss patterns do not follow average trends and highlights the need for both more research to understand individual weight loss facilitators and barriers, and to develop more personalized, culturally tailored and informed weight loss strategies.

Model 1: Low attendance cohort: Odds of being a paradoxical responder versus an expected non-responder a Model 2: High attendance cohort: Odds of being a paradoxical non-responder versus an expected responder b Adjusted odds ratio (95% CI) Adjusted odds ratio (95% CI)
a Model 1: Odds of response (≥5% weight loss; n = 415) versus expected outcome (<5% weight loss; n = 1083) among patients who attended less than 75% classes at 12 months follow-up.b Model 2: Odds of non-response (<5% weight loss; n = 140) versus expected outcome (≥5% weight loss; n = 175) among patients who attended 75% or more classes at 12 months follow-up.The largest contributor to the paradoxical response in the lowattendance cohort was non-Hispanic White race/ethnicity.Even after adjustment for various other sociodemographic and clinical factors, non-Hispanic White participants were almost twice as likely as Hispanic participants, 2.5 times as likely as non-Hispanic Asian participants, and almost 3 times as likely as non-Hispanic Black particineeds in the context of LMIs, and also the development of tailored LMIs that are culturally sensitive and able to respond to those needs.