Intervention Presentation on Diabetes

Systematic Review or Meta-analysis

Effectiveness of group-based self-management education

for individuals with Type 2 diabetes: a systematic review

with meta-analyses and meta-regression

K. Odgers-Jewell1 , L. E. Ball2, J. T. Kelly1, E. A. Isenring1, D. P. Reidlinger1 and R. Thomas3

1Faculty of Health Sciences and Medicine, Bond University, 2Menzies Health Institute Queensland, Griffith University and 3Centre for Research in Evidence-Based

Practice (CREBP), Bond University, Gold Coast, Australia

Accepted 17 February 2017

Abstract

Aims Patient education for the management of Type 2 diabetes can be delivered in various forms, with the goal of

promoting and supporting positive self-management behaviours. This systematic review aimed to determine the

effectiveness of group-based interventions compared with individual interventions or usual care for improving clinical,

lifestyle and psychosocial outcomes in people with Type 2 diabetes.

Methods Six electronic databases were searched. Group-based education programmes for adults with Type 2 diabetes

that measured glycated haemoglobin (HbA1c) and followed participants for ≥ 6 months were included. The primary
outcome was HbA1c, and secondary outcomes included fasting blood glucose, weight, body mass index, waist

circumference, blood pressure, blood lipid profiles, diabetes knowledge and self-efficacy.

Results Fifty-three publications describing 47 studies were included (n = 8533 participants). Greater reductions in
HbA1c occurred in group-based education compared with controls at 6–10 months [n = 30 studies; mean difference
(MD) = 3 mmol/mol (0.3%); 95% confidence interval (CI): �0.48, �0.15; P = 0.0002], 12–14 months [n = 27 studies;
MD = 4 mmol/mol (0.3%); 95% CI: �0.49, �0.17; P < 0.0001], 18 months [n = 3 studies; MD = 8 mmol/mol (0.7%); 95% CI: �1.26, �0.18; P = 0.009] and 36–48 months [n = 5 studies; MD = 10 mmol/mol (0.9%); 95% CI: �1.52, �0.34; P = 0.002], but not at 24 months. Outcomes also favoured group-based education for fasting blood glucose, body weight, waist circumference, triglyceride levels and diabetes knowledge, but not at all time points. Interventions facilitated by a single discipline, multidisciplinary teams or health professionals with peer supporters resulted in improved outcomes in HbA1c when compared with peer-led interventions. Conclusions Group-based education interventions are more effective than usual care, waiting list control and individual education at improving clinical, lifestyle and psychosocial outcomes in people with Type 2 diabetes. Diabet. Med. 34, 1027–1039 (2017) Introduction Patient education is an integral and vital component of successful diabetes care [1–3]. The main goal of diabetes patient education is to promote and support positive self- management behaviours to optimize metabolic control, improve long-term diabetes outcomes and quality of life (QOL), prevent complications, and reduce morbidity and mortality, while remaining cost-efficient [1,4]. Group-based education for individuals with Type 2 diabetes may be more cost-effective and efficient than individual education, due to the reduced time and funding required to educate numerous people in one sitting [5]. The potential advantages of group- based education interventions over individual visits include time for the provision of more detailed information, decreased time demands on health workers, easy incorpora- tion of families and carers, and facilitation of discussions and support from others facing the same challenges [6,7]. Clearly, the use of group-based education warrants further investigation. Three previous systematic reviews included group educa- tion for Type 2 diabetes. A Cochrane systematic review assessed the effects of group-based training on clinical, lifestyle and psychosocial outcomes in people with Type 2 diabetes compared with routine treatment, waiting list control or no intervention [8]. The review favoured group-Correspondence to: Kate Odgers-Jewell. E-mail: [email protected] ª 2017 Diabetes UK 1027 DIABETICMedicine DOI: 10.1111/dme.13340 http://orcid.org/0000-0002-1453-4447 http://orcid.org/0000-0002-1453-4447 http://orcid.org/0000-0002-1453-4447 based education, finding significant improvements in HbA1c levels, body weight, systolic blood pressure (BP), and fasting blood glucose (FBG), a decreased need for diabetes medica- tion and increased diabetes knowledge [8]. A subsequent publication in 2012, updating the original Cochrane review, supported the findings of the former, favouring group-based education, with significant reductions in HbA1c, FBG and body weight, and improvements in diabetes knowledge compared with controls [6]. Another recent systematic review [9] assessed the effect of diabetes self-management education and support methods, providers, duration and contact time on glycaemic control in adults diagnosed with Type 2 diabetes. The review included individual, group- based, combination and remote interventions for the man- agement of Type 2 diabetes, with results suggesting that a combination of individual and group-based education was most effective at improving HbA1c (median 9.6 mmol/mol; 0.88%) when compared with controls [9]. These previous reviews had limitations. First, the searches are outdated and the number of published studies for group- based diabetes interventions has increased substantially since their completion. High heterogeneity precluded meta-analyses for several of the main outcomes, which were completed for just two studies [6,8]. Although both reviews found clinical and statistically significant changes in health outcomes, the exact mechanism or ‘active ingredient(s)’ of these complex interventions were not identified [6,8]. Both reviews only conducted follow-up analyses of the primary outcome up to 2 years from baseline [6,8]. The quality of the previous reviews was assessed using ‘A Measurement Tool to Assess Systematic Reviews’ (AMSTAR), a reliable and valid method for assessing the methodological quality of systematic reviews [10]. The AMSTAR scores were categorized in line with previous research [11,12], with scores of 0–4 classified as ‘low quality’, 5–8 classified as ‘moderate quality’, and 9–11 classified as ‘high quality’. The Cochrane review [8] was assessed as a high- quality review (score: 9/11). This review lacked an assessment of publication bias and conflict of interest for the included studies. The review by Steinsbekk et al. [6] was assessed as a moderate quality review (score: 5/11); no protocol was available, grey literature and publication bias were not considered, a list of excluded studies was not provided, an assessment of conflict of interest for included studies was not explored and the scientific quality of the included studies was not used appropriately in formatting conclusions. The review by Chrvala et al. [9] was assessed as a moderate quality review (score: 7/11); grey literature and publication bias were not considered, a list of excluded studies was not provided and conflict of interest for included studies was not explored. The review had various limitations including: restricting included studies to English-language publications, including only ran- domized controlled trials, including interventions for individ- uals with either/both Type 1 and/or Type 2 diabetes, and an inability to conduct meta-analyses [9]. Despite these systematic reviews providing evidence of effectiveness, group-based education interventions are often complex and the characteristics of the interventions vary greatly, for example, in the number of contact hours, number of sessions, number and characteristics of participants, group facilitator(s) or educator (s) qualifications, facilitator training, theoretical framework, and whether family, friends or carers can attend [6,8]. Health professionals may deter from group- based education because the essential attributes for a success- ful group-based education programme are unknown. Fur- thermore, no specific evidence-based practice guidelines for group-based education in Type 2 diabetes have been identified internationally, inevitably resulting in wide variations in the programmes offered, and creating difficulty in the interpreta- tion of evidence and its translation to a practice setting. This systematic review builds upon two of the previous reviews [6,8] and seeks to update the evidence for the effectiveness of group-based interventions for Type 2 dia- betes management and investigate key attributes for success- ful group programmes. It was hypothesized that: � group-based interventions for Type 2 diabetes would have greater reductions in HbA1c compared with controls in the short (6 months) and long (> 12 months) term;

� group-based interventions for Type 2 diabetes would
improve body weight, body mass index (BMI), waist

circumference, FBG, BP, lipid profiles, diabetes knowledge

and self-efficacy, compared with controls;

� variations in effect sizes could be attributed to study design
(i.e. setting, control group, educator), and intervention

characteristics (i.e. number of participants, intervention

length, number of contact hours).

Methods

The study was registered with the International Prospective

Register of Systematic Reviews PROSPERO (CRD420150

27785).

What’s new?

• We present a comprehensive up-to-date review of the

evidence for the effectiveness of Type 2 diabetes group-

based interventions. This is the first review in the area

to complete a meta-regression.

• We report statistically significant results for improving

HbA1c, fasting blood glucose, body weight, waist

circumference, triglycerides and diabetes knowledge,

but clinical improvement is more nuanced.

• Group-based interventions facilitated by a single disci-

pline, multidisciplinary teams or health professionals

with peer supporters appear to be more effective at

improving HbA1c than peer-led interventions.

1028 ª 2017 Diabetes UK

DIABETICMedicine Group-based education in Type 2 diabetes � K. Odgers-Jewell et al.

Data sources and search strategy

A systematic literature search was performed to retrieve

publications on group-based education for the management

of Type 2 diabetes in adults. The search was completed in

three parts. First, electronic databases, including PubMed,

the Cochrane Central Register of Controlled Trials (CEN-

TRAL), Embase, CINAHL, PsycINFO and ERIC, were

searched from commencement of records to 22 September

2015 (File S1). Second, hand searches of reference lists from

previous reviews were completed [6,8]. Finally, the included

studies were cross-referenced with the results of an updated

search by the authors of the most recent review including

studies up to May 2012 (email correspondence). No

language or date restrictions were applied. Abstract-only

publications were excluded and duplicate articles were

removed prior to title and abstract screening.

Inclusion criteria and study selection

Group-based education intervention studies for participants

diagnosed with Type 2 diabetes that reported randomized

controlled trials, cluster randomized trial or controlled

clinical trial study designs were included. Studies were

included if the described intervention met the following

criteria: adults aged ≥ 18 years; face-to-face, educative
group-based interventions (including those with occasional

adjunct individual consultations) for people with Type 2

diabetes; a minimum of four participants and may include

family and friends for support; a minimum of one session

lasting for 1 h; groups delivered in primary or secondary care

compared with a control or comparison group (usual care,

waiting list control or individual intervention); and studies

that measured HbA1c at both baseline and 6 or more months

from baseline. Studies were excluded if participants were

pregnant women or were diagnosed with Type 1 diabetes, or

interventions provided education in individual consultations,

included only exercise prescriptions without education or

were not conducted face-to-face.

All studies were screened against the eligibility criteria by

two independent reviewers (KOJ and LEB) using reference

manager software EndNote (Thomson Reuters, New York,

NY, USA). Conflicts were resolved by discussion between

them. Studies that met the inclusion criteria or did not

include sufficient information for screening in the title and

abstract, were included for full-text review. Full-text versions

of these articles were obtained and screened independently.

Authors were contacted for missing data up to three times by

email if the missing data affected assessment of the study’s

eligibility, and were excluded if contact could not be made.

Data extraction and quality assessment

Data extraction was completed by the first author (KOJ) and

confirmed for accuracy by an independent reviewer (JTK).

Data extracted included: general information on the study

design, trial characteristics, intervention details, participant

characteristics, outcome measures, results and information

for appraising the risk of bias. Study quality was assessed

using the Cochrane risk of bias tool [13] by two independent

reviewers (KOJ and LEB). Disagreements were resolved

through discussion. Risk of bias was ranked as low, unclear

or high depending on whether a study had any element of

bias (e.g. selection, performance, detection, attrition, report-

ing and other bias).

Data synthesis and analysis

Descriptive data from the included studies were summarized.

Data were meta-analysed if the same measurement was used

across three or more studies at the same time point. The

primary outcome measure was change in HbA1c in group-

based education vs. control. The secondary outcome mea-

sures were changes in FBG, weight, BMI, waist circumfer-

ence, BP, total cholesterol, LDL-cholesterol, HDL-

cholesterol, triglycerides, diabetes knowledge and self-effi-

cacy. Prior to the meta-analyses, studies reporting FBG or

lipid profile measures in mg/dl were converted to mmol/l;

those reporting weight in lb were converted to kg.

Summaries of effect estimates were calculated by meta-

analysis using the DerSimonian and Laird random effects

model in Review Manager (RevMan, v. 5.3) [14]. Continu-

ous data using the same measures were analysed with a

weighted mean difference in outcomes between the interven-

tion and control groups, whereas continuous data collected

using a variety of measures were assessed using the standard

mean difference (SMD). Heterogeneity was assessed using

the I-squared statistic and reported following the Cochrane

Handbook [13].

Mean differences (MD) and 95% confidence intervals (CI)

were calculated in RevMan and standard error was calcu-

lated in Microsoft� Excel using the 95% CIs for the meta-

regression. Separate analyses for the effect of group-based

interventions on HbA1c were performed for the following

subgroups: control groups, delivery setting, insulin therapy,

type of educator(s), training of educator(s), baseline HbA1c
levels, theoretical model and intervention content, materials,

length, number of sessions, contact time, number of partic-

ipants and the inclusion/exclusion of family and/or friends.

Sensitivity analyses were conducted to explore the influ-

ence of study quality (overall risk of bias and reporting bias),

on HbA1c outcomes (as measured closest to intervention

completion) and heterogeneity. Reporting bias and selective

outcome reporting were chosen for the sensitivity analysis

because studies that did not report the pre-specified outcomes

or failed to include the results for an expected outcome may

be reporting only results supporting the studies’ aims or

hypotheses. We also examined potential influences on the

primary outcome for studies that had differences in HbA1c at

baseline, large (defined as > 10%) compared with small

ª 2017 Diabetes UK 1029

Systematic Review or Meta-analysis DIABETICMedicine

attrition (defined as < 10%), and studies published in non- English journals due to potential publication bias. Subgroup analyses were also conducted to examine the effects of different educators (health professionals, health professionals with peer support, peer or lay-persons), disci- plines (single discipline compared with multidiscipline) and studies that included participants taking (and not taking) insulin on the primary outcome HbA1c. In addition, a univariate meta-regression was completed to explore potential associations between the size of effect and varying study and intervention characteristics [15]. Variables were similar to those explored in the subgroup analyses. A meta-regression was performed using Stata statistical soft- ware [16]. Results Study selection The search identified 14 016 results, from which 9764 publications were screened against the selection criteria, leaving 298 studies for full-text review (Fig. 1). Forty-seven studies reported in 53 publications were included in the systematic review (references provided in File S2). A total of 8533 participants were included in the 47 studies (intervention group n = 4416, control group n = 4117). The mean age of participants was 60 years. Men made up 44% of participants in both the intervention (1917 of 4383) and control (1799 of 4086) groups. Three of the 47 included studies (6%) recruited only women. Known duration of diabetes was reported by 29 of the 47 studies (62%). Mean duration of diabetes was 8.9 years for participants in the intervention group, and 9.4 years in the control group. Mean HbA1c level at baseline was 67 mmol/mol (8.3%) for both groups and ranged between 39 and 111 mmol/mol (5.7%– 12.3%) for the intervention group and between 40 and 115 mmol/mol (5.8%–12.7%) for the control group. In 38 (81%) studies, the mean HbA1c was > 53 mmol/mol (7%)

for both the intervention and control groups.

Study characteristics

Study characteristics are detailed in Table S1. Of the 47

studies included, 40 reported the results of randomized

controlled trials, four reported results of controlled clinical

trials and three reported the results of cluster randomized

controlled trials. Most of the studies were carried out in the

USA (18; 38%), the UK (6; 13%) and Italy (5; 11%). Forty-

two of the studies were published in English, two in Spanish

[17,18], two in Italian [19,20] and one in Dutch [21]. The

studies were published between 1988 and 2015, and the

length of follow-up was 6–60 months from baseline.

Records identified through

Id
en
tif
ic
at
io
n

S
cr
ee
ni
ng

E
lig
ib
ili
ty

In
cl
ud
ed

Records after duplicates removed

database searching
(n = 14016)

(n = 9767)

Records screened

Full-text articles

Studies included in
quantitative synthesis

(meta-analysis)

assessed for eligibility

(n = 9767)
Records excluded

Full-text articles excluded

Did not meet intervention
criteria (n= 180)

Abstract only (n= 29)
Protocol/ commentary (n= 8)

Not an RCT (n= 34)

(n = 9469)

(n = 251):

(n = 298)

(n = 47)

Additional records identified
through other sources

(n = 3)

FIGURE 1 Stages of study identification

1030 ª 2017 Diabetes UK

DIABETICMedicine Group-based education in Type 2 diabetes � K. Odgers-Jewell et al.

Intervention characteristics varied in materials provided,

discipline(s) of group educators and theoretical model used,

as summarized in Table S2. Studies were conducted pre-

dominantly in primary care settings (32; 68%), with 15

(32%) of the studies delivered in secondary or tertiary care

settings, for example, hospital diabetes centres or tertiary

hospitals. Four publications [22–25] reported on multiple

arm studies.

Study quality

Most studies were classified as having a moderate (31) or

high (12) risk of bias, with four studies classified as having a

low risk of bias (Table S3). Of the six risk of bias items,

allocation concealment (selection bias), blinding of partici-

pants and personnel (performance bias), and blinding of

outcome assessment (detection bias) were the least consis-

tently described or were generally poorly conducted in the

included studies (Fig. S1).

Overall effects of group-based interventions for HbA1c

A meta-analysis was conducted to assess the effect of group-

based education compared with control for all 47 included

studies (n = 7055) using the measure of HbA1c at the time

point closest to the completion of each group-based educa-

tion intervention (Fig. 2). Overall, compared with control,

group-based intervention was effective in reducing HbA1c by

4 mmol/mol (0.3%) (95% CI: �0.51, �0.17; P < 0.0001; I2 = 84%). Heterogeneity was statistically significant and potential reasons for this were explored with sensitivity analyses. The results of the meta-analyses for HbA1c and secondary outcome measures at various time points are provided in Table 1. Group-based interventions significantly reduced HbA1c post intervention at most time points compared with controls. HbA1c was significantly reduced at 6–10 months post baseline [n = 30 studies; MD = 3 mmol/mol (0.3%); 95% CI: �0.48, �0.15; P = 0.0002; I2 = 65%], 12– 14 months post baseline [n = 27 studies; MD = 4 mmol/ mol (0.3%); 95% CI: �0.49, �0.17; P < 0.0001; I2 = 64%], 18 months [n = 3 studies; MD = 8 mmol/mol (0.7%); 95% CI: �1.26, �0.18; P = 0.009; I2 = 50%] and at 36– 48 months [n = 5 studies; MD = 10 mmol/mol (0.9%); 95% CI: �1.52, �0.34; P = 0.002; I2 = 93%]. By contrast, when eight studies comparing group-based interventions with controls measured HbA1c at 24 months post baseline, there was no significant difference between the groups. This time point also had the highest heterogeneity (I2 = 94%). There was variation in effectiveness in reducing FBG when comparing group-based interventions with controls. Group- based education was significantly more effective at reducing FBG compared with controls at 12–14 months post baseline (n = 8 studies; MD = 0.68 mmol/l; 95% CI: �1.25, �0.11; P = 0.02; I2 = 55%). However, this was not the case for FBG when measured at 6–10 or 24 months post baseline. All time points were assessed as having significant heterogeneity. Group-based education was significantly more effective at reducing body weight compared with controls at both 6– 10 months (n = 17 studies; MD = 1.22 kg; 95% CI: �2.22, �0.23; P = 0.02; I2 = 62%) and 12–14 months (n = 9 studies; MD = 1.43 kg; 95% CI: �2.09, �0.77; P < 0.0001; I2 = 0%). Despite the statistically significant improvements in body weight at two time points, group- based education was not effective at significantly reducing BMI. Group-based education was significantly more effective at reducing waist circumference at 6–10 months (n = 5 studies; MD = 1.19 cm; 95% CI: �2.34, �0.05; P = 0.04; I2 = 58%) than controls. However, although waist circum- ference showed a trend for improvement with group-based education at 12–14 months, the difference between groups was not significant (n = 3 studies; MD = 0.79 cm; 95% CI: �1.96, 0.38; P = 0.19; I2 = 38%). Both systolic and diastolic BP were measured at four time points (6–10, 12–14, 24 and 36–48 months post baseline). When pooled, changes in systolic and diastolic BP were not statistically different between groups for any of these intervals. In addition, there were no significant differences in total cholesterol between group-based interventions and controls at any time point. There were no statistically significant differences in HDL- cholesterol between groups at any time point. Heterogeneity was significant at all time points. There were mixed results for LDL-cholesterol when measured at two time points, 6– 10 and 12–14 months (Table 1). At 6–10 months, the meta-analysis resulted in no significant differences between groups for LDL-cholesterol (n = 12 studies; MD = 0.03 mmol/l; 95% CI: �0.13, 0.07; P = 0.59; I2 = 49%). The studies assessing LDL-cholesterol at 12– 14 months showed a significant decrease in LDL favouring the control group (n = 5 studies; MD = 0.08 mmol/l; 95% CI: 0.01, 0.15; P = 0.04; I2 = 0%). Group-based education was significantly more effective at reducing triglycerides at 6– 10 months (n = 14 studies; MD = 0.13 mmol/l; 95% CI: �0.24, �0.01; P = 0.03; I2 = 4%) and 24 months (n = 3 studies;MD = 0.32 mmol/l;95%CI:�0.58,�0.06;P = 0.01; I2 = 8%). At 12–14 months, the difference between groups for triglycerides were not significant (n = 11 studies; MD = 0.04; 95% CI: �0.22, 0.14; P = 0.66; I2 = 68%). Diabetes knowledge was reported by 16 studies and measured using a range of validated questionnaires. Group- based education was significantly associated with improved diabetes knowledge at two time points: 6–10 months (n = 7 studies; SMD = 0.61; 95% CI: 0.14, 1.08; P = 0.01; I2 = 83%) and 12–14 months (n = 7 studies; SMD = 0.58; 95% CI: 0.08, 0.97; P = 0.02; I2 = 93%). Self-efficacy was reported by five studies at three time points (6, 12 and 24 months). Group-based education showed a trend to improved self-efficacy at 12 months post baseline (n = 3 ª 2017 Diabetes UK 1031 Systematic Review or Meta-analysis DIABETICMedicine studies; SMD = 0.15; 95% CI: �0.02, 0.33; P = 0.08; I2 = 0%), however, these measures were not significant. In addition, QOL, depression, energy intake and physical activity levels were assessed (Table S4). Subgroup analyses Analyses were completed for 13 subgroups using HbA1c at the point closest to the end of each of the group-based education interventions as the outcome measure (Table 2). The type of educator subgroup analysis resulted in a significant subgroup difference (P = 0.002), with peer- and/ or lay-led group-based interventions having no significant influence on improving HbA1c (P = 0.80). Interventions facilitated by single disciplines (P = 0.0003), multidisciplinary teams (P = 0.02) or health professionals with peer supporters (P = 0.01) were effective at significantly improving HbA1c (Table 2 and Fig. S2). Types of educators were further analysed to individual disciplines included in the ‘single discipline’ group, finding that physician-, dietitian- and nurse-led group-based education interventions were effective (P < 0.00001) at improving HbA1c (Fig. S3). Heterogeneity for both subgroup analyses was significant (I2 = 79.1% and 89.2%, respectively). In addition, subgroup analysis of studies with regards to insulin therapy resulted in a significant subgroup difference (P < 0.00001), with inter- ventions that excluded participants on insulin therapy resulting in greater reductions in HbA1c (Fig. S4). Hetero- geneity for both subgroup analyses was significant (I2 = 69% and 88%, respectively). FIGURE 2 Effectiveness of group-based interventions compared with controls for Type 2 diabetes for HbA1c (%). Risk of bias: A, random sequence generation (selection bias); B, allocation concealment (selection bias); C, blinding of participants and personnel (performance bias); D, blinding of outcome assessment (detection bias); E, incomplete outcome data (attrition bias); F, selective reporting (reporting bias); G, other bias. 1032 ª 2017 Diabetes UK DIABETICMedicine Group-based education in Type 2 diabetes � K. Odgers-Jewell et al. Sensitivity analyses Sensitivity analyses were performed to explore the influence of study quality and characteristics on post-baseline HbA1c outcomes (Table 3) and heterogeneity. Forest plots for sensitivity analyses are reported in Figs S5–S9. There were no significant differences in HbA1c outcomes when study quality and attrition were explored; all subgroups showed statistically significant improvements in HbA1c (P ≤ 0.05). However, subgroups of studies assessed as being at high risk of reporting bias (Fig. S6), having baseline differences between groups (Fig. S7) and studies published in non-English journals (Fig. S9) did not show significant improvements in HbA1c. Meta-regression We used study variables and intervention characteristics including theoretical model, discipline(s) of educator(s), educator training, materials provided, delivery in primary care, both groups HbA1c < 7% at baseline, intervention length, contact time, number of participants, number of sessions, and the inclusion of family and friends in a meta- regression to explore potential associations between the size of effect and study and intervention characteristics on HbA1c at post intervention. None of these variables explained significant portions of heterogeneity among the studies (Table S5). Table 1 Summary of meta-analysis results for primary and secondary outcome measures at various time points Outcome Time point (months) No. studies No. participants (IG/CG) Mean difference (95% CI) P-value Heterogeneity (%) (I2) Heterogeneity (P-value) HbA1c (%) 6–10 30 2155/1952 �0.31 (�0.48, �0.15) 0.0002 65 < 0.00001 12–14 27 2233/2151 �0.33 (�0.49, �0.17) < 0.0001 64 < 0.00001 18 3 98/96 …