Results
A summary of search findings is shown in figure 1. Of 9058 studies found in initial searches, we included 132 studies. All the included studies are described in online supplemental table 1 and referenced in the reference list; however, not all are cited in the text of this paper. Disagreement during the abstract screen was 2.45%. There was no disagreement on the full-text screen. We found studies reporting on the following: anthropometrics (n=128), cardiovascular complications (n=27), BMD (n=12), nutritional deficiencies (n=34), puberty and menstruation (n=13) and other complications that did not fit into the previous categories (n=14). We found 63 cross-sectional studies, 8 longitudinal, 9 research trials (baseline data reported only) and 55 case studies/series. Studies originated from the USA, Canada, Brazil, Japan, Indonesia, Australia, Turkey and Europe. The majority (n=105) of studies were exclusively on children and adolescents ≤18 years, only 15 of which were exclusively on children ≤5 years. The remaining 27 studies were mixed samples of CYP (≤25 years).
Figure 1Flow diagram of searches and study selection. ARFID, avoidant restrictive food intake disorder; ED, eating disorder.
Quality of studies
Quality assessments are shown in online supplemental table 1. Median case completeness of case studies was 75% (range: 36%–93%). Median NOS score in observational studies were: case-controls: 6 out of 9* (range: 4–9*); cohorts: 6 out of 9* (range: 4–8*); cross-sectional: 6 out of 10* (range: 4–9*). We also found five out of the nine research trials had moderate risk of bias; the remaining four had high risk of bias.
Anthropometrics
We found 128 studies reporting anthropometric data (online supplemental table 1), including 15 studies which compared HC8–23 and 28 which compared AN with ARFID.8–10 14 16 19 23–45
Seven of the 13 studies reporting weight distribution of CYP with ARFID found the majority were living with underweight (defined within the papers as <5th BMI centile or ≤−2 weight z-score),15 18 46–50 while six found the majority were within a healthy weight range (defined within the papers as 5th to 85th BMI centile or −2 to 1 weight z-score).51–56
We found six studies to pool in meta-analysis comparing BMI z-score or %MBMI in ARFID to HC (figure 2), producing a large overall effect size for lower standardised BMI scores in ARFID (SMD: −0.99; 95% CI: −1.40 to –0.58; n=846)13–15 18 21 23 compared with HC. Heterogeneity was high (I2: 79.81%). All pooled studies were in children ≤18 years old, except one study on CYP aged 10–23 years13 (removal had no significant impact on effect size or heterogeneity). All studies were clinical samples, except one which was a community sample and was the only study of ‘poor’ quality rather than ‘good’ quality as assessed by the NOS21 (removal increased the effect size (SMD: −1.14; 95% CI: −1.46 to –0.82; n=558) and reduced the heterogeneity, I2: 56.95%.
Figure 2Forest plot showing standardised mean difference in BMI z-scores and %Median BMI between ARFID (n=228) and HC (n=618). 95% CIs and study weights and indicated. The overall effect size was calculated using a random effects model. ARFID, avoidant restrictive food intake disorder; BMI, body mass index; HC, healthy control.
We also found 15 studies to pool for meta-analysis comparing BMI z-score or %MBMI in ARFID versus AN, producing no overall difference in effect size (SMD: −0.00; 95% CI: −0.31, 0.31; n=1689; figure 3)9 14 16 19 23 25 27 31 33 34 36 38–40 44; heterogeneity was high (I2: 86.57%). Of the 15 studies included, 11 were of children ≤18 years old,14 16 19 23 25 31 33 34 38 40 44 and the remaining four studies included CYP <25 years of age9 16 27 36 39 (the removal of which did not affect effect size or heterogeneity). All studies were in clinical samples, except two16 23 (removal of which did not affect effect size or heterogeneity). Most studies were of ‘good’ quality; however, six were of ‘fair’ quality.19 31 33 38 40 44 Removal of ‘fair’ quality studies19 31 33 38 40 44 increased the difference between ARFID and AN; however, SMD was still insignificant (SMD: 0.18; 95% CI: −0.19 to 0.54; n=1004) and heterogeneity was still high (I2: 82.85%). A funnel plot was not perceived to display asymmetry, confirmed with Egger’s test (p=0.64). Meta-regression using available data on age and sex across studies found no significant associations with effect size (data not shown).
Figure 3Forest plot showing standardised mean difference in BMI z-scores and %Median BMI between ARFID (n=597) and AN (n=1092). 95% CIs and study weights and indicated. The overall effect size was calculated using a random effects model. AN, anorexia nervosa; ARFID, avoidant restrictive food intake disorder; BMI, body mass index.
Most of the longitudinal anthropometric data we found came from case studies (n=14),57–70 most reporting rapid weight loss in the months before presentation. We found three larger studies reporting longitudinal data.39 46 71 One study found that individuals with ARFID had lower pre-diagnosis BMIs and lost significantly less weight during their illness than patients with AN (%MBMI lost: 15% vs 21%; p=0.03).39 In addition, another study reported individuals with ARFID lost a mean of 9.6±9.1 kg before presentation for treatment.71 The only study that followed individuals with ARFID for multiple years found that the percentage of them that had severe acute malnutrition (weight z-score ≤−2; from 76% at 2 years of age to 52% at 11 years of age) and those who had severe chronic malnutrition (height z-score ≤−2, 51% to 25%) declined over the years.46 Five cross-sectional studies mentioned growth faltering or stunting in ARFID; the reported prevalence of growth delay across all five studies ranged from 1.4% to 51%.46 51 52 54 55
Cardiovascular complications
We found 27 studies containing data on cardiovascular parameters in ARFID (online supplemental table 1).8 19 24 32 33 35 39 41 58 60 61 63 67 70–82 One compared ARFID to HC8 and seven compared ARFID to AN.8 19 24 33 35 39 41
All non-case studies that contained data on HR or blood pressure (BP) in individuals with ARFID found the mean HR and BP to be within a normal range for age.8 19 24 33 35 41 75 78 82 Several studies did, however, report some individuals with ARFID (3.92%–52.6%) had bradycardia, and around 2% had hypotension.32 33 39 70 73–75 Five case studies reported tachycardia in ARFID58 67 72 80 83; most were reported in the context of electrolyte abnormalities or severe nutritional deficiencies. Two studies investigated potential risk factors for variance in HR and BP within the ARFID population.75 77 One study identified a subgroup within ARFID that had rapid weight loss and a shorter length of illness also had a lower HR.77 In contrast, another study found individuals with ARFID who had acute compared with chronic ED symptom onset had similar HRs and BPs (acute: 76.2±15.5 beats per minute (bpm), chronic: 79.4±15.1 bpm, NS).75 The weight status of the ARFID participants also had no significant impact on their HR or BP, but CYP <12 years had significantly higher HRs (82.40±17.84 bpm vs 76.15±12.23 bpm; p<0.05) and lower systolic blood pressure (109.72±10.14 vs 115.61±8.31; p<0.01) than CYP ≥12 years.
All studies comparing ARFID to AN, reported an average higher HRs, lower prevalence of bradycardia and higher BP in ARFID, despite similar BMIs.8 19 24 33 35 39 We found five studies to pool in a meta-analysis comparing HR in ARFID to AN, showing that mean HR was greater in ARFID (WMD: 12.93 bpm; 95% CI: 8.65 to 17.21; n=685; figure 4).8 19 24 33 35 Heterogeneity was high (I2: 81.33%). All studies included were clinical samples, with two studies of ‘good’ quality8 24 and three of ‘fair’ quality.19 33 35 Only one study included had CYP over 18 years old (10–22 years old)8; removal of which increased the difference in HR between ARFID and AN (WMD: 15.5 bpm; 95% CI: 12.87 to 18.19; n=623). All studies used retrospective medical data apart from two, which were cross-sectional studies8 35 which when removed increased the mean difference in HR between ARFID and AN to 16.20 bpm (95% CI: 13.28 to 19.13; n=517) and reduced the heterogeneity to null (I2: 0.00). There was no difference in mean standardised BMI between ARFID and AN in all studies included, except for one19 were individuals with AN had a lower standardised BMI; removal of which slightly reduced the mean difference and heterogeneity (WMD: 11.69; 95% CI: 7.08 to 16.31; n=398; I2: 75.91%). Meta-regression and subgroup analysis found no moderators to this relationship (eg, age, percentage female, type of study, year published).
Figure 4Forest plot showing weighted mean difference in heart rates (beats per minute) between ARFID (n=283) and AN (n=402). 95% CIs and study weights and indicated. The overall effect size was calculated using a random effects model. AN, anorexia nervosa; ARFID, avoidant restrictive food intake disorder.
Bone mineral density
We found 12 studies that reported data on BMD within ARFID (online supplemental table 1)22 24 25 59 84–88; eight of which were case studies,59 70 80 81 84–87 one compared BMD in ARFID to HC22 and two compared BMD in ARFID to AN.24 25
In the case studies/series that stated BMD z-score, all individuals with ARFID, except for two,85 86 had BMD z-scores ≤−2 (BMD z-score range: spine 0.4 to –4.1; hip −3.1 to –4.6).59 81 85 87 Furthermore, one cross-sectional study reported that 25% of individuals with ARFID had BMD z-scores ≤−2 in their spine while 77% had BMD z-scores ≤−1.24 Another study found that the mean BMD z-score in individuals with ARFID was −2.46.88 Only one study mentioned Bone Mineral Density Apparent Density (BMAD) z-scores, where the mean BMAD z-score in individuals with ARFID was −1.44.25 Lower BMI, lower BMI z-score, amenorrhea and delayed puberty were associated with low BMD.25 BMI and BMI z-scores were associated with BMAD z-scores. Despite a lower BMI being a risk factor for low BMD, two cases of severe osteoporosis were reported in teenage boys with healthy weight (hip BMD z-scores: −4.1 to −4.6).59 87
In the one study that compared BMD z-scores in ARFID to HC, individuals with ARFID had significantly lower BMD z-scores in Total body (−1.41 vs −0.5; p=0.021) and total body less head (−1.67 vs −0.74; p=0.055); however, in the lumbar spine the difference not significant (−0.95 vs −0.67).22
Two studies compared BMD z-scores in ARFID to AN. One study found that individuals with ARFID had lower BMD z-scores in their lumbar spine than those with AN (−2.00 vs −1.38, ARFID vs AN, p<0.001).24 Another study reported that although scores were low in both ARFID and AN (−1.88±0.91 and −1.43±1.18, respectively), there was no significant difference between the two.25
Nutritional deficiency
We found 34 studies mentioning micronutrient levels in ARFID (online supplemental table 1).15 43 50 52–54 57–59 69 70 72 80–100 However, only two studies compared dietary nutritional content in ARFID to HC15 94 and one study to AN.96
Five studies contained detailed data on the dietary nutritional content in ARFID; all of which reported a high prevalence of individuals with ARFID not eating the daily recommended intake (DRI) of several micronutrients.15 52 53 94 96 One study found that 67% of individuals with ARFID consumed <80% of the DRI of six or more micronutrients.52 Furthermore, our searches identified 22 individuals with ARFID suffering from severe micronutritional deficiencies which led to clinical disorders: xeropthalmia due to vitamin A deficiency86 87 90 99; nutritional optic neuropathy due to vitamin B12 and folate deficiency87 91; Wernicke encephalopathy due to vitamin B12 deficiency43; severe osteoporosis due to vitamin D and/or vitamin B12 deficiencies59 80 81 84 87; scurvy due to vitamin C deficiency52 70 80 84; pulmonary artery hypertension from vitamin C deficiency70; rickets due to vitamin D deficiency81 and iron deficiency anaemia.72 80 83 90 91 93 95 97
In comparison to HC, a higher percentage of individuals with ARFID were not meeting the DRI and were consuming a lower percentage of the DRI in almost all micronutrients tested.15 94 In contrast, individuals with ARFID consumed less of some micronutrients, such as vitamin C and A, than those with AN but more of others, such as selenium and magnesium.96
Puberty and menstruation
We identified 13 studies containing data on puberty and menstruation in ARFID (online supplemental table 1).8 9 22 29 33 36 38 62 63 67 75 76 101 Six studies compared pubertal and menstrual measurements in ARFID to AN,8 9 29 33 36 38 and three compared them to HC.8 9 22
Amenorrhea (primary and secondary) prevalence was reported at around 10% in females with ARFID.33 75 Where menstruation data in women with ARFID were compared with those with AN, all studies found significantly more individuals with AN experienced problems with their menstrual cycles, such as a higher prevalence of amenorrhea and more irregular periods, despite similar BMIs.8 29 33 36
Pubertal data were reported in three studies; however, all three included participants recruited from the same pool. One study reported that age at menarche in ARFID was older than in HC (13.1 and 12.7 years, respectively); however, this was not statistically significant.8 In contrast, age at menarche in AN was similar to ARFID (13.2 years). Two of the three studies reported that individuals with ARFID were at significantly lower Tanner stages in breast and pubic hair development than AN and HC.8 9 However, the individuals with ARFID were significantly younger than the HC, and those with AN. Therefore, the differences in Tanner stage and age at menarche were most likely due to the age difference. In contrast, the final study, which contained individuals with ARFID who were of similar ages to the HCs, found that they were at similar breast and pubic hair tanner stages to those HCs.22
Other physical complications
We found 21 studies that reported physical health complications associated with ARFID, which did not fit into the previously described categories (online supplemental table 1).32 39 58 60 61 67 72–74 97 98 102 103
The prevalence of electrolyte abnormalities in ARFID was reported to be between 23.1% and 73.7%,32 39 73 with one study reporting that 23.1% of individuals with ARFID had hypokalaemia and 7.7% had hypophosphatemia.32 Other electrolyte abnormalities mentioned across studies were hypochloraemia and elevated bicarbonate.32 58 67 72 Furthermore, in one case of ARFID in a 3-year-old boy, his hypokalaemia was so severe it led to rhabdomyolysis.58 Another study found that significantly more individuals with ARFID had electrolyte abnormalities than those with AN (23% and 10% respectively, p=0.03).39
Several other common physical complications associated with low weight and EDs were reported in ARFID, such as lanugo, lethargy, dizziness, syncope, pale skin, muscle wasting, dehydration, cognitive problems, headaches, hypothermia, diarrhoea and constipation.60 61 72 76 97 98 One study found that 93 out of 207 (44.9%) CYP with ARFID had one or more of these medical symptoms associated with their ED.76 Furthermore, some severe, but rare, conditions associated with ARFID, such as Ogilvie’s syndrome, hypoalbuminemia and superior mesenteric artery syndrome, were reported by included studies, but the prevalence rates remain unknown.60 67 72 74