Papers Flashcards
Everest and Emery
Data Visualization Checklist for High Impact
- left justified text
- horizontal text
- accurate proportions and equidistant axis intervals
- 2 dimensional graph
- readable with color blindness
- muted gridlines or none
- no border
Mayo and Hand
Replication crisis of p-values due to misuse of statistical tools
Statistical significance tests address error due to mistaking random variability for an effect
Saying “statistically significant” means the observed difference is not readily explained by chance - not scientific importance
P value does not measure effect size
P-value is a probability
A statistically non-significant result does not prove the null hypothesis
P-hacking increases type I error probability from selective reporting
P-value is NOT the probability that a test or null hypothesis is true, it is the probability that the observed results are due to chance
Harding et al.
Induced colitis model used to study effect of adequate nutrition vs malnutrition (normal in children with colitis) on growth rates plus probiotics
Never before tested effect of probiotic therapy on nutrition status as measured by weight gain and protein synthesis
32 piglets, 4 groups: healthy and well-nourished, colitis and well-nourished, colitis and malnutrition and colitis, malnutrition and probiotics
Repeated measures ANOVA + Dunnett’s post hoc test, Mann Whitney U test for immunoblotting results
Conclusion was that well nourished pigs grew better but probiotics did not help
Gaisawat et al.
Are cells resistant to C difficile inflammation when treated with probiotics ?
Obtained fecal sample with C difficile and healthy fecal sample: 8 different treatments
Cytokine production assessed with two-way ANOVA vs fecal water control + Tukey’s post hoc
Adding time in created 3 factors total: strain, treatment or no, time
2 were significant R0011 and SB between 0-24 hours, but only SB was significantly different from control
One-way ANOVA for cell viability used for each fecal treatment + Dunnett’s post hoc vs control
Would be too difficult to run a 3-way ANOVA so they excluded normal vs. CDI and only showed the CDI (ANOVA starts after red line)
Dunnett’s posthoc test to compare to control
Cytokine analysis done with two-way ANOVA followed by Tukey’s posthoc
Yue et al.
Purpose of study was to verify validity (by spearman’s correlation) and reproducibility (by intraclass correlation coefficient) of SFFQ to estimate flavonoid and subclasses of flavonoid and flavonoid rich food intakes (7DDR as comparison)
FFQ is a convenient method of intake questionnaire, but 7DDR is more accurate
- FFQ completed twice to assess reproducibility
Sample size was n = 641 men and 724 women from men’s and women’s lifestyle validation studies, all health professionals, all white
Validity of SFFQ determined by mean flavonoid intake, and variance in SFFQ vs 7DDR
Results were stratified by sex because source data was already stratified and to observe differences in intake
Subclasses not well capture by SFFQ: onions, flavones (high in onions) and flavonols
General conclusions: SFFQ has good validity and reproducibility despite some considerations
Considerations:
Flavone estimates varied greatly depending on database and intake differences but overall effect on stud results was minimal due to low flavones overall
Isoflavones intake was very skewed - may not be generalizable
Goetz et al.
Relationship between sleep and body composition in pre-school children
Statistical method: multivariate linear regression
Primary dependent variables: body fat (DXA) and fat mass
Independent variables: sleep variables, sleep duration, midpoints
Covariates: such as PA, meal timing, energy intake, energy expenditure with accelerometer and doubly labelled water, BMI
Baseline (cross sectional assessment) and then 1 year later (longitudinal)
Conclusion: sleep midpoint and percent body fat and fat mass were significantly related (see asterisk)
- For every 1 hour increase in time sleep midpoint there was a 0.92% body fat increase - later bed time associated with higher body fat percentage
Criticism: all 3 sleep variables were added together instead of separating them in analysis - multicollinearity concern
Calleja et al.
RCT: Dairy has been found to be useful for weight management (fat, protein, vitamins) - from observational studies
3 study groups: Recommended dairy (24), low dairy (25) and control (14, 2007 food guide)
- Control group no intervention due to change of height at that age
Sample size calculation: determined beforehand that 22 participants required at 95% power to look at lean mass change between 2 groups (intervention groups 2x size of control)
Used intention to treat and last observation carried forward (instead of imputation)
primary outcome: body composition
secondary outcome: fitness
Statistical tests used: repeated measure analysis of covariance which accounts for the effect of time and group effect
Covariate: body weight - used simple randomization so covariate use allowed for even distribution of body weights amongst groups
Conclusions: recommended dairy + exercise can reduce body fat while increasing lean mass without body weight change w/o caloric deficit
Bhat et al.
Systematic review and meta-analysis on food prescription program: provide access to healthy foods as an intervention in chronic disease risk factors
- no health care workers included because they were focused on patients
- provided with vouchers or directly given food
13 studies included
5 studies only fruit and veg intake servings %
Inverse-variance weighted random effects meta-analysis
Percent change = change from baseline, absolute change = net change
Results were that food prescriptions did improve outcomes but results to be taken with caution due to grades based on Cochrane’s risk of bias tool and GRADE
Seibold et al.
To determine if 2019 longitudinal studies they found in PLOS ONE journal were reproducible
Groups of 2-3 master’s students worked together and tried to reproduce the statistical analysis as stated in the methods
11 out of 14 provided open access
Able to reproduce 6 out of 11 of the papers
Barriers: raw data instead of cleaned data, code not available, complexity of study, insufficient data description
