Attrition and Retention in Multidisciplinary Weight Management Interventions for Adults with Obesity: A Systematic Review


 Background

High rates of attrition undermine the success of weight management interventions (WMIs), but a comprehensive understanding of the factors that increase dropout risk remains absent. This is partly explained by heterogeneity of intervention design, and the absence of a universal definition of attrition. This systematic review aimed to identify the factors related to- and predictive of attrition and retention in multidisciplinary WMIs for adults with obesity.
Methods

The systematic literature search, conducted in Cochrane, Medline, PsycInfo, and Scopus, aimed to identify original research articles published between February 2008 and December 2019. Articles investigating attrition or retention in multidisciplinary WMIs were eligible for inclusion if interventions were for adults (≥18) with obesity identified by body mass index ≥30kg/M2 and lasted ≥6 months. Multidisciplinary was defined as ≥2 interventionist disciplines or professions, for the purpose of this review. Data was synthesised narratively.
Results

The literature search resulted in seventeen studies which satisfied the inclusion criteria. Attrition rates ranged from 10% at 3-months to 81% at 3-years. The sociodemographic factors associated with reduced risk of attrition included older age, living in less deprived areas, higher levels of education, and female gender. Poor mental health, low social support, high weight loss goals and poor or unsatisfactory results may increase the likelihood of participant dropout, but evidence was limited and inconclusive because of different methodologies, and only a small number of studies investigating some of the variables.
Conclusions

The scope for targeted retention strategies is limited because few variables were consistently associated with attrition. Until a comprehensive understanding of attrition emerges, WMIs should seek to reduce social inequities in the benefit of WMI provision. Future research should consider factors reported qualitatively, such as intervention expectations and satisfaction, social support, patient-clinician relationships, and logistical barriers. Adopting a universal definition of attrition and de-homogenising participant dropouts would advance future research. As qualitative evidence is limited, exploring participant experiences of WMIs would help understand how attrition rates can be reduced, and in-turn improve WMI effectiveness.

At present there is limited evidence of effective retention strategies in WMIs. One meta-analysis reported nancial incentives, a multicomponent intervention, and the inclusion of self-monitoring technology successfully reduced attrition in WMIs [20]. A systematic review of 63 WMIs concluded it was di cult to compare attrition across interventions because of design variations, including exclusion criteria, intervention methods, de nitions of attrition and length of follow-up [1]. To overcome these limitations, the current review focused on multidisciplinary (MD) non-surgical WMIs for adults with obesity (body mass index (BMI) ≥30kg/M2), designed to provide a minimum of 6-months treatment. This was decided because MD WMIs are best equipped to address the many aspects of obesity [21,22], and 6-months is considered a realistic duration to achieve clinically signi cant weight loss [22]. The aim of this systematic review was to identify the factors related to-and predictive of attrition and retention in MD WMIs for adults with obesity.

Methods
The review method followed the Preferred Reporting in Systematic Reviews and Meta-analyses guidelines [23], and was registered with PROSPERO; study ID: CRD42018110067.

Literature search
The databases searched were; Cochrane, Medline, PsycInfo, and Scopus. In addition, a call was sent to members of the Association for the Study of Obesitya UK organisation consisting of individuals working in the eld of obesity, for any relevant articles, and titles in reference lists not identi ed in the literature search were also included. The search terms used were; attrition OR dropout OR drop out OR retention OR retain OR attendance OR adher* AND diet OR weight management OR weight reduction OR weight control OR weight loss OR lifestyle OR multi-disciplinary service OR multi-disciplinary treatment OR obesity treatment.

Study collection and data synthesis
A rst database search was conducted on February 19th and 20th 2018 to identify studies published since 2008, and a second was performed on January 27th 2020, to identify studies published between February 2018 and December 2019. Eligibility criteria are shown in Table 1. Multidisciplinary treatments typically include a team of professionals, however, the scarcity of publications from MD obesity services has been highlighted elsewhere [24], consequently for the purpose of this review, interventions which employed at least 2 different interventionists of different disciplines/professions were considered su cient.
Where it was unclear whether articles met the inclusion criteria, studies were discussed between authors, and corresponding authors were contacted when necessary. Interventions had to be multidisciplinary (≥2 interventionist disciplines), designed to provide treatment for at least 6-months Studies must have investigated attrition/retention quantitatively using inferential statistics or qualitatively Descriptive characteristics include baseline sample, study design, eligibility criteria, intervention methods, follow-up timeframe, attrition rate, de nition of attrition/completion employed, and all statistical results. Due to the heterogeneity in study characteristics including design, intervention methods, de nitions of-and timepoints at which attrition was measured, data was synthesised narratively. For this review, factors associated with attrition and retention were categorised according to seven thematic categories; sociodemographic, weight/shape factors, behavioural, physical health, psychological health, personality & character, practice & intervention characteristics, similar to previous reviews [1,25].
Quality assessment.
Studies were assessed for methodological quality using the Checklist for Measuring Quality [26], which is recommended for assessments of randomised and non-randomised studies [27,28,29]. The question used to assess statistical power was amended to represent appropriate statistical analyses. Studies were judged on whether a multivariate regression analysis was performed, and to what extent the variation in the outcomes could be explained by the regression model. A sample of four studies were assessed by authors JE and EDB, and results were compared and discussed to identify any inconsistencies, which informed the assessment of the remaining studies.

Results
In total, 8482 studies were retrieved through database searches, and 19 through other sources. Following removal of duplicates, 5416 records were screened, resulting in 77 full texts assessed for inclusion, of which 17 were considered to have satis ed the inclusion criteria. An overview of all included studies is given in Table 2. The complete data identi cation process is shown in gure 1.

Statistical analyses & qualitative follow-up
All studies performed inferential statistical analyses to identify differences between completers and dropouts, using either null hypothesis tests or univariate regression analyses. All but three studies [11,30,31] followed these analyses with varying forms of multivariate regression. One study reported the model prediction rate [32], but none reported the level of variation explained by the model (R 2 equivalent). In addition to quantitative analyses, some studies used qualitative methods to explore post-treatment reasons for dropout (summarised in Table 3). One study undertook telephone interviews [17], and another performed a focus group [33]. Three studies collected participant-reported reasons for dropout but did not state the methods employed [10,34,35].
One study showed completers were more likely to be born in the country providing the intervention but did not perform multivariate analyses [31]. Another reported black-were less likely to attend than white participants but only in univariate analyses [33]. A third study reported 2-year retention was higher in white participants but not at 6-months, and ethnicity did not predict retention following multivariate analyses [34]. One study showed English rst language predicted 1-year-but not 3-year retention [32].
In three studies the odds of WMI completion were higher in individuals from least deprived areas [4,36,38], one of which identi ed higher early retention (de ned as attending >2 appointments) in participants from the least deprived areas [4]. One reported the relationship was not present in a female-only analysis [38], and another reported no difference in deprivation indices across longer-term retention (>12 months) [4]. There was no relationship with attrition when anxiety and depression was strati ed by deprivation indices in one study [11].
Marital status was not related to attrition [10,17,33,34]. One study reported odds of continued engagement were lower for those with children in the household, but family structure was not associated with attrition, and focus group participants reported family issues was the reason for dropout [33]. One study reported no relationship between the number of children in the household and attrition [17]. In another study, an unstated number of participants reported dropout was a result of personal or family reasons [35]. A study from the USA reported health insurance with a speci c provider predicted 2-year retention, but 96% of participants had insurance with this provider, and 25 of 30 dropouts reported change in health-insurance was the reason for dropout [34].

Body weight factors
Two studies reported completers had a lower baseline BMI [17,34], but only one showed this following multivariate analysis [34]. Two studies reported higher BMI predicted completion [36, 38], but the results were not repeated in a male-only analysis in one [38]. A fth study showed BMI was a predictor of retention at 1-year [32] but did not report the direction of the relationship (whether higher or lower BMI predicted retention). Eight showed no relationship [4,30,31,35,37,39,40,41].
Four of the six studies which considered the relationship between baseline weight and attrition found no relationship [30,35,39,40], and two studies reported odds of dropout increased with higher baseline weight [17,42]. Three studies found no relationship between waist circumference and attrition [30,35,40], and hip-circumference and waist-hip ratio was also unrelated in one [40]. A Croatian study reported a signi cant relationship between completion and lower baseline waist circumference in univariate but not multivariate analyses [17]. One study found participants with higher body fat percentage had lower odds of completion [40], and one reported no relationship [30].
Of weight-related history factors considered, one study reported completers had a history of greater weight cycling but did not support this with multivariate analyses [31]. According to one study, childhood obesity predicted retention at 1-and 3-years [32]. Two studies reported younger age at rst diet attempt predicted dropout [35,40], but in one of the studies this was only evident in participants who received cognitive behavioural therapy (CBT), and not controls (no CBT) [35]. Three studies showed no relationship between age at onset of obesity and attrition [17,31,41]. Lowest or highest adult weight was not associated with attrition in two studies [35,40], and one showed the number of diet attempts in cases and controls (with or without CBT), was not related to attrition [35]. Two studies reported no relationship between parental obesity and attrition [17,31].
Two studies considered weight loss goals, one showed no relationship [40], and one reported odds of dropout increased with 'acceptable' and 'disappointing'-, but not 'dream' or 'happy' weight loss goals [42]. Dropout was also associated with weight loss required to achieve weight loss goals in the latter. Of two studies investigating weight loss results, one found lower weight loss at 1-month predicted dropout [40], and the other found no signi cant relationship [31]. An unstated number of participants in one study reported unsatisfactory weight loss as a reason for dropout [35].
Impact of weight on quality of life (IWQOL)-lite scale was not associated with attrition [34], neither was weighing at least monthly [31]. Completion was associated with less body dissatisfaction [31], but a multivariate analysis was not performed to support this nding.

Behaviours -dietary & other
One study found completers reported lower baseline energy intakes in a univariate analysis [30], and one reported no relationship [33]. Macronutrient intake [30], bre intake [30,33], diet type (Mediterranean vs standard) [17], and fruit and vegetable intake [33] were not signi cantly related to attrition. Of the diet and eating behaviour questionnaires explored, one study showed dropouts scored higher on the three-factor eating questionnaire hunger scale, but they did not perform a multivariate analysis [30], and another reported higher scores on the eating assessment test (EAT-26) scale predicted attrition in a univariate-but not multivariate logistic regression [39]. Other factors unrelated to attrition included: three-factor eating questionnaire restraint and disinhibition scales [30], EAT-26 total score, food preoccupation and oral control subscales [39], eating disorder [31,41], binge eating [31,35,37,40] and night-eating [31,37].
One study reported non-smoking predicted 3-but not 1-year retention [32], and two reported no relationship [17,33]. Drinking alcohol was unrelated to attrition in one [33].

Physical health
A range of comorbidities were considered in relation to attrition. Of studies which considered cardiovascular diseases, one reported patients with hyperlipidaemia were more likely to attend >6 months in univariate-but not multivariate analyses [4]. One reported that having coronary artery disease predicted retention at 1-and 3-years [32], two showed no relationship between attrition and heart disease [4,38], and one found no relationship with baseline cholesterol levels [30]. Dropouts had lower baseline diastolic blood pressure in one study, but only in univariate-and not multivariate analyses [40].
Hypertension predicted retention at 1-and 3-years in one study [32], and in another patients with hypertension were more likely to attend >6 months in univariate-but not multivariate analyses [4]. Attrition was not related to baseline blood pressure in two studies [30,32], systolic blood pressure in one [40], and hypertension in another [38].
Of the studies that considered diabetes, HbA1C or blood glucose levels, three reported no relationship [30,32,38], and one reported patients with diabetes were more likely to attend >6 months in univariate-but not multivariate analyses [4].  Thyroid issues (hypothyroidism and thyroid problems) was not associated with attrition [17,38]. In one study, female patients with chronic obstructive pulmonary disease and stroke were less likely to complete [38]. One study reported patients with sleep apnoea were more likely to attend >6 months [4], and another showed presence of asthma predicted 3-but not 1-year retention [32]. Attrition was not related to osteoarthritis [38], and baseline comorbidities in four studies [10,33,34,39].
In one study, patients with joint pain were more likely to attend >6 months in a univariate-but not multivariate analyses [4]. One study identi ed no relationship between attrition and health-related quality of life metrics using Euro quality of life 5 dimensions (EQ-5D) mobility, including self-care, usual activity or pain problems [34]. In one study a small number of participants said health-related issues caused dropout [17], and medical reasons (non-obesity related surgery) were reported by 15 of 46 dropouts in the 'Healthy Weights Initiative' [10]. In another study, an unstated number of participants said physical limitations or health issues were the reasons for dropout [33]. Three studies reported no relationship between medication use and attrition [17,33,38].

Psychological health
Of six studies that considered the relationship between anxiety and attrition, one reported fewer patients with anxiety completed at 6-months, and fewer patients with severe anxiety completed at 3-6-and 12-months, but they did not perform a multivariate analysis [11]. They also reported no relationship between completion and anxiety when strati ed by gender. Five studies reported no relationship [34,35,38,39,41].
Of studies that considered depression or depressive symptoms, one reported non-completion was associated with moderate or severe depressed mood, but not after a multivariate analysis [10]. One study reported fewer patients with depression completed at 6-and 12-months, and fewer patients with severe depression completed at 12-months, but a multivariate analysis was not performed to support the ndings [11]. The same study reported no relationship between depression and 3-month completion, and severe depression and 3-and 6-month completion. One study reported completion at 2-years was associated with less depressive symptoms using the inventory of depressive symptomology-short revised (IDS-SR) in univariate-but not multivariate analyses [34]. They also reported no relationship between depression and 6-month and 2-year retention, or between inventory of depressive symptomology & 6-month attrition. Six studies reported no relationship between depression and attrition [4,17,35,38,39,40].
One study reported non-completion was associated with greater psychiatric disturbances (using symptom checklist revised, global severity index (SCL-90-R-GSI), irritable mood, type A behaviour (using diagnostic criteria for psychosomatic research (DCPR) framework) and having >1 DCPR diagnoses, but not following multivariate analysis [39]. They also reported no relationship between attrition and demoralisation or illness denial (DCPR diagnoses). One study reported dropout was associated with presence of any Axis I disorder, but not following multivariate analysis, but retention was predicted by absence of any mental disorder [41]. They also reported no relationship between attrition and mood disorder, somatoform disorder or impulse-control disorder. Neither mental distress [31] or mental illness were related to attrition [32]. An unstated number of participants said stress caused dropout in one study [33].

Personality/character
Of studies that investigated the temperament and character inventory (TCI) scale and attrition, one reported completion was associated with higher harm avoidance and reward-dependence, but only the latter was signi cant after multivariate analysis [41]. Novelty seeking, reward dependence, persistence, selfdirectedness, co-operativeness, self-transcendence was not associated with attrition. Another study reported that dropouts scored higher for novelty-seeking and lower for self-directedness in univariate-but not multivariate analyses, and harm avoidance, persistence, reward-dependence, co-operativeness and selftranscendence were unrelated to attrition [37].
Higher scores on the anger-hostility subscale of the SCL-90 predicted dropout in two studies [35,40], the latter was speci c to controls (participants who did not receive CBT). Both reported no relationship between attrition and other SCL-90 subscales [35,40].
The only study to consider alexithymia reported it predicted dropout [39]. Retention at 2-years was associated with better quality of life using the IWQOL-lite questionnaire in one study, but not following multivariate analysis [34]. One study reported no relationship between attrition and quality of life measured with SF-36, physical component summary and mental component summary scales [30]. Others showed no relationship between attrition and weight locus of control [31], or stage of change [33].
The sole study to investigate attrition and motivation for weight loss and to change habits reported no signi cant relationship [31], but lack of motivation was reported as a reason for dropout by participants in two studies [17,35]. Participants without a signed 'buddy social support contract' were less likely to complete in one study [10], and an unstated number of participants said lack of social support was a reason for dropout in another [33].

Practice or intervention characteristics
In one study, odds of completion were greater in practices with a patient list size of 4000-8000 compared to <4000, but lower in practices with a higher percentage of their population from deprived areas (>40% vs <15%) [36]. Being a training practice, patient list size >8000, practice rating (quality and outcome framework points), practice population with 15-40% of its patients from the most deprived areas, and practice referral rate were not related to attrition [36]. Travel distance was not related to attrition in two studies [34,36]. Three of thirty participants reported that moving area was the reason for dropping out in one study [34]. One study reported that cases (receiving CBT) were more likely to complete [33], and a small number of participants said treatment dissatisfaction caused dropout in another study [17].

Quality assessment
Studies were mostly of moderate quality, as overall quality scores ranged between 13 and 20 (median 18) of a maximum score of 32. This was mostly due to the observational designs employed, which particularly limited the scores for internal validity and confounding. The median score and range for each section was; 7/11 (5-9) for reporting, 2/3 (1-3) for external validity, 4/7 (3)(4)(5) for internal validity, 4/6 (2-4) for confounding, and 1/5 (0-1) for power. No study reported the R 2 equivalent for the regression models, limiting scores for power.

Discussion
The aim of this systematic review was to identify the factors associated with-and predictive of attrition and retention in MD weight management interventions for adults with obesity.
Participant sociodemographic characteristics were the most consistently explored variables in the studies included in this review. Older age was often associated with reduced risk of dropout, which may be explained by differences in weight loss motives and preferences for WMI content between younger and older adults. Previous research shows older adults are more likely to cite health concerns as their motive for weight loss, and their efforts are more likely to be triggered by medical reasons [43]. Qualitative insight also suggests weight loss efforts are often motivated by concern from family members, and major family events such as the birth of children and grandchildren [44]. Participants also reported health-driven motives stem from the desire to avoid ill-health or ameliorate existing ill-health, as opposed to the acquisition of better long-term health, which may be more pertinent in older age.
There was some evidence that men were more likely to dropout, however most studies reported no difference between sexes. It has been shown that women utilise primary care services more than their male counterparts [45], and increased interactions with healthcare services may partly explain why women are sometimes more likely to continue engaging than men. A series of systematic reviews of male obesity management concluded that tailoring WMIs can bene t men, for example, men prefer WMIs set in the community include physical activity and nutrition [46]. They also like to focus on health as opposed to appearance and want regular contact with a high degree of personalisation, as well as setting goals, and incorporating humour and camaraderie.
This review identi ed evidence that participants from deprived areas were more likely to dropout, in addition to participants with lower levels of education.
Previous research has shown individuals from deprived areas are less likely to participate in some health services [47,48] and more likely to miss appointments [49], but the reasons why are not well understood. Participants of WMIs who are more socioeconomically deprived may face nancial constraints which could undermine autonomy, requiring them to make compromises which their more a uent counterparts do not, compounding weight management challenges [50]. Furthermore, a review of socioeconomic inequalities of health care in England showed that, although people from deprived areas typically utilise healthcare services more than their richer counterparts, socially advantaged people are more likely to use preventive healthcare services and seek medical support at earlier stages of illness [51]. Such differences in the way healthcare services are used may help explain the relationship between attrition and deprivation. Ethnicity-related factors were scarcely investigated, and variables were different across studies, limiting the identi cation of trends related to attrition.
Some evidence suggests baseline weight measures were related to attrition, but the direction of the relationship was inconsistent, and the majority of studies reported no relationship. There was some quantitative and qualitative evidence that weight loss goals and results were related to attrition, but this was limited due to the small number of studies which investigated it. Similarly, a narrative review reported high weight loss expectations had often been associated with increased risk of dropout, but not in all studies [52]. This could be due to whether or not WMI expectations are addressed within the intervention, or qualitative differences in the reasons why participants engage in WMIs, which is not exclusively driven by weight loss. Some participants may focus on improving health, gaining support or developing new habits [53], which may offset unsatisfactory weight loss.
There were no trends evident between differences in dietary intakes or behaviours and attrition. There was limited evidence that smoking was related to attrition, and no evidence that alcohol intake affected participation, however these were scarcely explored.
There were some signi cant relationships between obesity-related comorbidities and attrition, but there was a limited number of studies which considered the same comorbidities. The majority of studies which investigated anxiety or depression (including depressive symptoms), reported no relationship with attrition, however, a large UK-based NHS service did report that presence of anxiety and severe anxiety, and depression and severe depression was related with shortand longer-term attrition. Poorer mental health has been associated with attrition in previous research [1], and inconsistent results may be partly explained by the different questionnaires used to assess mental health.
A number of psychometric scales were employed to explore aspects of personality and character and their relationship with attrition, but there was no consistent evidence of any relationship. However, only a small number of studies considered such factors, and they were smaller scale studies, limiting the available evidence. Signi cant relationships were reported between presence of alexithymia, a personality construct characterised by a de cit in emotional processing, i.e. an inability to explain and differentiate between emotions [54].
There was no evidence that travel distance signi cantly contributed to attrition, but only two studies considered this. In contrast, a qualitative study previously reported travel distance was a reason for dropout by one-fth of 766 participants [15], demonstrating logistical barriers such as travel may vary depending on the geographical area.

Strengths of this review
There was a need to review studies investigating attrition in similarly designed WMIs, which this review has ful lled by only including WMI studies which were MD and designed to treat adult obesity for a minimum of 6 months. This evidence gap was present because previous reviews had not restricted their inclusion criteria to similarly designed WMIs, and the broad range of interventions limited the extent to which results could be compared [1]. This review aids the understanding of attrition speci c to longer-term MD WMIs for the treatment of adult obesity. This review is relevant because only studies published within the previous 12 years were eligible for inclusion, purposely set to identify WMIs with methods which were re ective of current weight management practice.

Limitations
Most of the evidence comes from observational research, and the absence of control groups means inferences are di cult to make, however, value is gained from exploring attrition in 'real world' services. There is a lack of publications from longer-term MD WMIs reporting on attrition, demonstrated by the fact few studies satis ed the inclusion criteria, which re ects the limitations of a previous systematic review of MD WMIs [24].
Further limitation stems from the different de nitions of attrition or completion employed across studies. De nitions were based on either losing contact, loss to follow-up, non-attendance or level of attendance, and some failed to clearly de ne attrition. This is a key limitation because even amongst comparable WMIs, it cannot be known whether contentious results stem from inconsistent de nitions or valid differences between studies. In addition, this review was limited by the fact that despite a large number of variables having been studied, variables such as treatment expectations, treatment results, logistics and measures of personality and character were only explored in a small number of studies, which were mostly smaller scale studies, limiting the sum of evidence available to draw conclusions.

Recommendations
Weight management services should focus on reducing social inequities in the bene t of WMI provision, as a comprehensive understanding of attrition remains elusive and current evidence suggests sociodemographic factors are the most likely to in uence dropout.
In line with the 'Star Lite' core outcome measures, WMIs should report on attendance and completion rates, the number of participants who drop out and reasons why [55]. The Consolidated Standards of Reporting Trials statement also advise this [56]. Future publications should clearly de ne attrition and/or completion, and the need for a universal de nition of attrition remains. This could be de ned in accordance with the core outcome measures, which includes the measurement of; attendance at core sessions, completion (attend >80% of core sessions), and dropout/non-completion (participants who attend <80%). Studies may also bene t from de-homogenising dropouts into early ( rst 1/3 of treatment) and late ( nal 2/3 of treatment), as recommended elsewhere [3].
This would aid comparisons across studies and help develop a more comprehensive and nuanced understanding of attrition.
Studies should explore variables including participant expectations and satisfaction, the patient-clinician relationship, social support, and logistical barriers, which are reported in qualitative research [15], but scarcely explored in the studies in this and a previous review [1]. Qualitative research into attrition is needed, and we support a previous call for studies to explore participant experiences to help improve the design and acceptability of WMIs, and ultimately their effectiveness [14]. Ubiquitous technologies (e.g. short-messaging service), by minimising participant burden and nancial cost, can facilitate large-scale qualitative data collection in WMIs [57], which could be utilised in future research.

Conclusion
The studies reviewed suggest older age, living in less deprived areas, higher levels of education and female gender were the sociodemographic factors with evidence of reduced risk of attrition. There was some, but limited evidence that poor mental health, low social support, high weight loss goals and poor or unsatisfactory results may increase the likelihood of dropout. Until a comprehensive understanding of attrition is developed, WMIs should focus on reducing social inequities in the bene t of WMI provision. Future research should investigate participant expectations and satisfaction, patient-clinician relationships, social support, and logistical barriers, which are reported qualitatively but scarcely investigated quantitatively. Studies should ensure attrition/retention is clearly de ned, and a universally recognised de nition would aid comparisons. This could be de ned as attendance or non-attendance at 80% of the core The authors declare that they have no con icts of interest.

Funding
This work is part of a PhD funded by the Knowledge Economy Skills Scholarship 2 (KESS 2) awarded to JE.
Authorship JE developed the protocol, conducted the literature search, data extraction, critical appraisal, and data synthesis, and wrote the draft and nal version of the manuscript.
DH assisted assessment of whether studies satis ed the inclusion criteria and commented on drafts of the article.
EB-D assisted the development of the protocol, literature search, assessment of whether studies satis ed the inclusion criteria, critical appraisal, and commented on drafts of the article.
AS assisted the development of the protocol, literature search and assessment of whether studies satis ed the inclusion criteria and commented on drafts of the article.
All authors approved the nal version of the manuscript.