Eating disorders and social media use in the OxWell Student Survey

Mina Fazel

Department of Psychiatry, University of Oxford, United Kigdom

Objectives

  • This session will provide an overview of associations between eating difficulties and social media use among young people and emerging adults from a large population-based study: the OxWell 2023 Student Survey.

Methods (I)

  • The OxWell Student Survey is a repeated cross-sectional online survey that asks students (aged 8-18 years) in England over 200 questions on a range of mental health and wellbeing measures.

  • The 2023 survey has two different versions according to year group (divided into English school years 5 to 6, and 7 to 13/further education). This presentation will share data from students in school years 7 onwards (i.e. secondary school students).

  • The study was approved by the University of Oxford Research Ethics Committee (Ref: R62366/R002).

Methods (II)

  • We use an adapted version of DAWBA(Moya et al. 2005), where questions are answered “Yes” or “No”. -

  • Loneliness has been developed by Geulayov et al. (2022), and data was reduced to an index using Multiple Correspondence Analysis.

DAWBA adapted
X2060 Have you ever thought you were fat even when other people told you that you were very thin?
X2070 Would you be ashamed if other people knew how much you eat?
X2080 Have you ever deliberately made yourself vomit (throw up)?
X2090 Do worries about eating (what? where? how much?) really interfere with your life?
X2100 If you eat too much, do you blame yourself a lot?
X2110 Do you regularly skip meals at school because you have shape and/or weight concerns?

Methods (III)

  • We model multinomial multivariate logistic regressions adjusting by age, gender, ethnicity, neurodivergency, care and SES status.

  • We adjusted p-values due to multiple comparisons.

Descriptive statistics

Characteristic N = 23,135
Gender
Boy 10,247 (45%)
GD 316 (1.4%)
Girl 11,705 (51%)
GND 666 (2.9%)
Missing 201
DAWBA
One positive answer 3,730 (16%)
Two positive answers 2,534 (11%)
Three or more positive answers 8,342 (36%)
Negative 8,529 (37%)
Self-harm 3,846 (17%)
Missing 600
Loneliness -0.28 (-0.60, 0.08)
Sleep trouble
Not at all 7,117 (31%)
A little 8,375 (36%)
Somewhat 3,912 (17%)
Much 2,195 (9.5%)
Very much 1,314 (5.7%)
NoResponse 222 (1.0%)
NoResponse(Stopped_vA) 0 (0%)
Anxiety and depression (RCDAS) 7,370 (32%)
Maltreatment -0.04 (-0.05, 0.02)
1 n (%); Median (IQR)

  • 23,135 students included in the main analysis.

  • Sample was reduced to those that answered all 6 questions of DAWBA adapted.

  • 63% of respondents answered at least yes to one of the DAWBA adapted questions.

Multinomial models (I)

(1)
(2)
Characteristic Outcome N OR p-value Outcome N OR p-value
Anxiety and depression (RCDAS) Two positive answers 22,029
No Two positive answers 14,978 —
Yes Two positive answers 7,051 1.64 <0.001
Anxiety and depression (RCDAS) Three or more positive answers 22,029
No Three or more positive answers 14,978 —
Yes Three or more positive answers 7,051 4.30 <0.001
Anxiety and depression (RCDAS) Negative 22,029
No Negative 14,978 —
Yes Negative 7,051 0.56 <0.001
No. Obs. 22,029 22,029
Loneliness Two positive answers 22,029 1.54 <0.001
Loneliness Three or more positive answers 22,029 3.27 <0.001
Loneliness Negative 22,029 0.54 <0.001
1 OR = Odds Ratio

Note: Models are adjusted by age, gender, ethnicity, neurodivergency, care and SES status

The reference group (r.g.) for all models is “One positive answer”

Model 1 looks at the association between anxiety/depression (as measured by the RCDAS) and the DAWBA eating disorder assessment outcomes.

  • Compared to those without anxiety/depression:
    • Those with anxiety/depression had 1.64 times higher odds of having two positive DAWBA answers in comparison to the reference group (p<0.001)
    • They had 4.30 times higher odds of having three or more positive DAWBA answers (compared to r.g.) (p<0.001)
    • They had 0.56 times lower odds of having a negative DAWBA assessment (compared to r.g.) (p<0.001)

Model 2 model examines the relationship between loneliness and the DAWBA eating disorder outcomes.

  • Compared to those with lower loneliness:
    • Those with higher loneliness had 1.54 times higher odds of having two positive DAWBA answers (compared to r.g.) (p<0.001)
    • They had 3.27 times higher odds of having three or more positive DAWBA answers (compared to r.g.) (p<0.001)
    • They had 0.54 times lower odds of having a negative DAWBA assessment (compared to r.g.) (p<0.001)

Multinomial models (II)

(3)
(4)
Characteristic Outcome N OR p-value Outcome N OR p-value
Self-harm Two positive answers 21,489
No Two positive answers 17,843 —
Yes Two positive answers 3,646 1.59 <0.001
Self-harm Three or more positive answers 21,489
No Three or more positive answers 17,843 —
Yes Three or more positive answers 3,646 3.99 <0.001
Self-harm Negative 21,489
No Negative 17,843 —
Yes Negative 3,646 0.54 <0.001
No. Obs. 21,489 22,029
Maltreatment Two positive answers 22,029 2.73 <0.001
Maltreatment Three or more positive answers 22,029 9.34 <0.001
Maltreatment Negative 22,029 0.21 <0.001
1 OR = Odds Ratio

Note: Models are adjusted by age, gender, ethnicity, neurodivergency, care and SES status

Model 3 looks at the association between being classified as self-harmer and the DAWBA eating disorder assessment outcomes.

  • Compared to those not self-harming:
    • Those self-harming had 1.59 times higher odds of having two positive DAWBA answers (compared to r.g.) (p<0.001)
    • They had 3.99 times higher odds of having three or more positive DAWBA answers (compared to r.g.) (p<0.001)
    • They had 0.54 times lower odds of having a negative DAWBA assessment (compared to r.g.) (p<0.001)

Model 4 model examines the relationship between loneliness and the DAWBA eating disorder outcomes.

  • Compared to those with lower maltreatment:
    • Those with higher loneliness had 2.73 times higher odds of having two positive DAWBA answers (compared to r.g.) (p<0.001)
    • They had 9.34 times higher odds of having three or more positive DAWBA answers (compared to r.g.) (p<0.001)
    • They had 0.21 times lower odds of having a negative DAWBA assessment (compared to r.g.) (p<0.001)

Multinomial models (III)

Outcome Characteristic N OR p-value
Two positive answers Sleep troubled 21,866
Not at all 6,808 —
A little 8,032 1.19 0.007
Somewhat 3,712 1.42 <0.001
Much 2,109 1.76 <0.001
Very much 1,205 1.19 0.3
Three or more positive answers Sleep troubled 21,866
Not at all 6,808 —
A little 8,032 1.52 <0.001
Somewhat 3,712 2.31 <0.001
Much 2,109 3.50 <0.001
Very much 1,205 4.32 <0.001
Negative Sleep troubled 21,866
Not at all 6,808 —
A little 8,032 0.62 <0.001
Somewhat 3,712 0.53 <0.001
Much 2,109 0.53 <0.001
Very much 1,205 0.59 <0.001
No. Obs. 21,866
1 OR = Odds Ratio

Note: Models are adjusted by age, gender, ethnicity, neurodivergency, care and SES status

This model looks at the association between self-reported quality of sleep and the DAWBA eating disorder assessment outcomes.

  • Any level of sleep trouble were associated with increased odds of having more positive answers on the DAWBA. The odds ratios were higher for the “Three or more positive answers” category than the “Two positive answers” category across all levels of sleep trouble.

  • This suggests the existence of a stimulus-response relationship between severity of sleep trouble and the likelihood of having more positive answers on the DAWBA assessment. However, there is no directionality suggested here.

Multinomial models (IV)

Outcome Characteristic N OR p-value
Two positive answers Hours_spent 17,744
0 hrs 1,778 —
30 mins 1,767 0.85 0.2
1 hr 1,944 0.98 0.9
1hr 30 mins 0 1.00 >0.9
2 hrs 2,517 1.12 0.4
3 hrs 2,569 1.17 0.2
4 hrs 2,102 1.23 0.12
5 hrs 1,558 1.11 0.5
6 hrs 1,102 1.32 0.074
7 hrs 548 1.16 0.5
8 hrs or more 1,859 1.21 0.2
Three or more positive answers Hours_spent 17,744
0 hrs 1,778 —
30 mins 1,767 0.91 0.4
1 hr 1,944 0.98 0.8
1hr 30 mins 0 1.00
2 hrs 2,517 1.31 0.009
3 hrs 2,569 1.55 <0.001
4 hrs 2,102 1.80 <0.001
5 hrs 1,558 1.97 <0.001
6 hrs 1,102 2.03 <0.001
7 hrs 548 1.97 <0.001
8 hrs or more 1,859 2.20 <0.001
Negative Hours_spent 17,744
0 hrs 1,778 —
30 mins 1,767 0.78 0.008
1 hr 1,944 0.74 0.001
1hr 30 mins 0 1.00 >0.9
2 hrs 2,517 0.72 <0.001
3 hrs 2,569 0.69 <0.001
4 hrs 2,102 0.64 <0.001
5 hrs 1,558 0.70 <0.001
6 hrs 1,102 0.60 <0.001
7 hrs 548 0.72 0.036
8 hrs or more 1,859 0.65 <0.001
No. Obs. 17,744
1 OR = Odds Ratio

Note: Models are adjusted by age, gender, ethnicity, neurodivergency, care and SES status

This model looks at the association between time spent in the social media and the DAWBA eating disorder assessment outcomes.

  • Odds ratio are positive for the association of 3 hours or more and the “Three or more positive answers” category using as a reference group “One positive answer” in DAWBA (p.<.001). There seems to be a stimulus-response relationship, where odds increase progressively with more hours spent. Compared to the reference, odds start becoming significantly higher at 2 hrs (OR=1.31) and continue increasing up to 8 hrs or more (OR=2.20).

  • In an opposite direction, spending more hours per day on the internet is associated with lower odds of getting a “Negative” DAWBA rating, which implies higher odds of having some level of eating disorder problem detected by the DAWBA assessment.

Linear models (V)

Characteristic N Beta p-value
Hours_spent 18,588
0 hrs 1,871 —
30 mins 1,833 0.07 0.021
1 hr 2,016 0.11 <0.001
2 hrs 2,620 0.17 <0.001
3 hrs 2,713 0.24 <0.001
4 hrs 2,208 0.29 <0.001
5 hrs 1,648 0.28 <0.001
6 hrs 1,153 0.31 <0.001
7 hrs 570 0.27 <0.001
8 hrs or more 1,956 0.33 <0.001
No. Obs. 18,588

Note: Models are adjusted by age, gender, ethnicity, neurodivergency, care and SES status

This model looks at the association between time spent in the social media and how much the respondent worries about body shape and/or weight (Likert scale with 5 options transformed to continuous variable).

  • There is a positive association between hours spent in the internet and worrying about body shape and/or weight (p.<.001). The effect size is greater for more more hours in the internet, reaching a .33 standard deviation (effect size considered “medium”) increase on the scale of worrying about body shape and/or weight when respondent spends 8 hours or more in the internet (in comparison to the category 0 hrs).

Multinomial models (VI)

These models look at the association between access to a social media platforms (16 questions) and the DAWBA eating disorder assessment outcomes.

  • After adjusting for multiple comparisons, 9 out of 16 models show a statistically significant associations between DAWBA - three or mode positive answers” in comparison to “One positive answer” and use of social platforms (p.<.001) (see axis X in plot).

  • Odds Ratio of displaying “Three positive answers or more” in DAWBA ranges from 1.3 to 1.9 in comparison to “One positive answer” for the following platforms: Omegle, Pinterest, BeReal, Facebook, Instagram, Twitter, and Snapchat.

  • We computed between 25% to 35% of probabilities of “Three positive answers or more” in DAWBA according to different platforms.

  • There are not significant associations between DAWBA - “Two positive answers” in comparison to “One positive answer” and use of social platforms.

Multinomial models (VII)

These models look at the type of content accessed in the internet (7 options) and the DAWBA eating disorder assessment outcomes.

  • After adjusting for multiple comparisons, 5 out of 7 models show a statistically significant associations between DAWBA - three or mode positive answers” in comparison to “One positive answer” and the type of content accessed in the internet (p.<.001) (see axis X in plot).

  • Odds Ratio of displaying “Three positive answers or more” in DAWBA in comparison to “One positive answer” ranges from 1.3 to 1.7 in comparison to those not reporting access to that type of content.

  • We computed ~ 30% to 33% of probabilities of “Three positive answers or more” in DAWBA according to different platforms

  • There are no significant associations between DAWBA - “Two positive answers” in comparison to “One positive answer” and type of content accessed.

Conclusions

  • Models suggests that anxiety and depression, higher levels of loneliness self-harm, maltreatment and sleep trouble are statistically associated with having at least two disordered eating symptoms. We adjust p-values to account for a multiple comparisons problem.

  • Social media platforms, time spent in the internet and certain types of content accessed in the internet are also associate with having at least two disordered eating symptoms

References

Geulayov, Galit, Karen Mansfield, Christoph Jindra, Keith Hawton, and Mina Fazel. 2022. “Loneliness and Self-Harm in Adolescents During the First National COVID-19 Lockdown: Results from a Survey of 10,000 Secondary School Pupils in England.” Current Psychology 43 (15): 14063–74. https://doi.org/10.1007/s12144-022-03651-5.
Moya, Tatiana, Bacy Fleitlich-Bilyk, Robert Goodman, Fabiana Chamelet Nogueira, Patrícia Souza Focchi, Manoela Nicoletti, Vanessa Pinzon, Táki Athanássios Cordás, and Francisco Lotufo Neto. 2005. “The Eating Disorders Section of the Development and Well-Being Assessment (DAWBA): Development and Validation.” Revista Brasileira de Psiquiatria 27 (1): 25–31. https://doi.org/10.1590/s1516-44462005000100008.