EPM202 FA 2011-2012. Feedback sheet:

Student:

Grade awarded:

General comments:

i) Main strengths and weaknesses of this FA

ii) Grade

iii) Specific areas where this FA could have been improved/suggestions for areas for further study/review

iv) Response to student queries in FA cover sheet

Name of Marker:

General note to students:

Thank you for submitting this FA. By working through the analysis you should have been able to apply the concepts and methods that you have studied in EPM202. This feedback grid can be read alongside the suggested answer for the assignment, and the Stata dofile and commands used to produce the “solution” in the suggested answer, both of which you should also receive along with this feedback.

Please note the comment at the start of the suggested answer about other possible approaches to this analysis. This is very important. Although relevant sections from the suggested answer are highlighted here in this feedback it is to help you to identify some useful descriptions and appropriate sections in the text, however your FA will be marked on its own merits, using the way you have chosen to recode and use the data, and which may therefore be different to those suggested below, and on your description and justification of your methods, and your presentation and interpretation of the results you have obtained. An excellent answer will not necessarily be the same as the sample answer.

There will also be an Elluminate session organised for after all students have received their feedback for the FA, but before the AA deadline, which will go through and explain some of the analysis strategies and choices that you could have made. We will be very happy to answer any questions that you may have about this assignment then, or at any stage via the web-board.

Feedback grid:

Criteria / % marks awarded for section / Answered well? (Yes, somewhat, No) / Relevant text from suggested answer / Relevant/possible Stata commands - the attached do-file is a more comprehensive resource here / Comments
Introduction – Should provide an overview of the study including the following points:
a) Description of the study / 5 / In this study, dietary intake over a week was measured in 337 people. They were followed up for incidence of coronary heart disease (CHD). The aim of the analysis in this report is to find out whether there is an association between energy intake (EI) and rates of incident CHD. / stset dox, fail(chd) id(id) /*
*/ origin(dob) enter(doe) /*
*/ scale(365.25)
Note:when you use the stset command the output includes a basic description of your data set
b) A description of the numbers of CHD events (46) and proportion (46/337=14%) / 3 / Of the 337 in the study, 46 (14%) individuals were diagnosed with CHD. / tab _d (variable _d created when “stset” is used)
c) Description of follow-up time (3 months-20 years) / 2 / follow‐up times ranged from about 3.5 months to 20 years. / gen days_in_study=_t-_t0
sum days_in_study
Statistical methods and analysis strategies: Should describe the analysis strategies taken for the study
d) Statistical methods used / 5 / Distributions of data describing the study participants and their diets were examined and tabulated.
Association of supplied characteristics with CHD incidence was tested by calculating rate ratios for categories of the variables, and testing for statistical significance of devation of the rate ratios from one. Significant predictors of CHD were also tested (using chi-square test) for statistical association with EI: those predictors which were also associated with EI were considered potential confounders of the EI/CHD relationship. The EI/CHD association was then stratified by potential confounders. If no significant heterogeneity was detected among strata a pooled Mantel-Haenszel rate ratio for CHD incidence comparing EI levels was produced. This was compared with the crude rate ratio to assess the extent of confounding. / Example: distribution of age of study participants at the start of the study
hist agein, bin(20) addlab saving(hagein, replace)
e) Missing values -
Comment about the issue and make an appropriate attempt to address it. / 5 / There were 5 missing values for height, and 4 each for weight and fibre (different people). Since these numbers of subjects with missing values were low (nine (2.7%) in total) and excluding them would not be likely to introduce bias it was decided to conduct the multivariate analyses only on those with complete data for any variables in the analysis. Univariate analyses were conducted using all data available. / tab height, missing
includes missing values when the data is tabulated
f) Data manipulation -
Correct creation of a new binary variable for energy intake: low energy/ higher energy using the median (2802 kcal/d) and correct description / 5 / To simplify analyses, continuous variables energy, fat and fibre intake, height and weight were each categorised using the median to create two categories, called “low” and “high”, respectively. Age at entry was grouped mostly into 5-year categories. / * to calculate the median
centile energy, c(50)
(exposures)
g) Recoding of other risk factors (should state cut-offs used and some reason for their choice) / 5 / As above. Note cut-offs should also be stated when results and descriptive statistics are tabulated
Results – univariate analyses
h) Crude analysis – main exposure (EI)
Correct crude rate ratio for energy intake (EI) reported: high energy intake seems to reduce the rate of CHD / 5 / Table 1, 1st 2 lines
i) Comment on statistical significance
Mention and correct interpretation of statistical significance of crude RR for EI / 5 / Significant or marginally significant at 5% level: P=0.045 for n=337. Table 1, line 2 / stmh
j) Univariate analyses including stratified description other exposures
Number of cases, person-years by EI and other risk factors / 5 / Table 1, suggested answer / strate
k) Univariate analyses, main exposure and other possible confounders/effect modifiers
Rate/rate ratio by EI and other risk factors - the reader should be able to calculate the rates for categories if they are not reported as such / 5 / Table 1, suggested answer / strate (stratified rates)
stmh (rate ratio)
l) Statistical test
P-value or 95% CI by EI and other risk factors / 5 / See results in table 1
m) Description of results
Mention of the predictive effects of other risk factors (in tables or text) / 5 / The oldest age‐group at the start of the study suffered a significantly higher rate of CHD than the youngest age group. Month of survey did not predict later CHD rate (data not shown). High EI was highly significantly related to high fat intake (χ2 (1df)=431; P<0.001), high fibre intake (χ2 (1df)=187; P<0.001), greater height (χ2 (1df)=8; P=0.004) and greater weight (χ2 (1df)=40; P<0.001). EI was also significantly associated with age at start of study in a non‐linear pattern ((χ2 (5df)=18; P=0.003). Job was not significant.
Results - Stratified Analysis Ref: CAL SM02 – looking for confounding and interaction
n) Some mention of purpose of stratified analysis / 5 / The EI/CHD association was then stratified by potential confounders. If no significant heterogeneity was detected among strata a pooled Mantel-Haenszel rate ratio for CHD incidence comparing EI levels was produced. This was compared with the crude rate ratio to assess the extent of confounding. / stmh energy2 by(height2)
example, results stratified by height
o) Confounding
Well-justified choice of confounders used in adjusted results / 5 / Analysing only people with complete data, the marginally significant crude relationship between EI and CHD rate (RR=0.57; P=0.068) was made slightly less significant and the RR increased towards one by single adjustment for height, weight or age group at start (Table 2). Adjustment for fibre or especially for fat completely removed any hint of statistical significance of the crude relationship. There were no statistically significant interactions between EI and the adjusting variables. Thus these variables all appeared to be confounders to varying extents, so that after adjustment we have not found evidence for any significant relationship between EI at the start of the study and later rates of CHD in this population.
Comments correctly about interaction when interpreting the M-H results
MH-stratified and pooled RR for EI and 95% CI or P-value in the Tables or Figures / 5 / Table 2. “There were no statistically significant interactions between EI and the adjusting variables.”
p) Mention and correct interpretation of statistical significance of adjusted results / 5 / As above. “ Adjustment for fibre or especially for fat completely removed any hint of statistical significance of the crude relationship”
Discussion and conclusion
q) Conclusion about the results, especially a mention of the effect of energy variable after controlling for other risk factors / 10 / Correct conclusion: Crude relationship seen between CHD and EI was not robust.
The crude relationship between EI and later CHD was only marginally significant. Given the long follow-up times it would be rather surprising to find a robust relationship between them. Adjusting for fat or fibre intakes could be questioned, especially in the case of fat intakes, since EI is highly correlated with each one. It is not likely that fat could be considered to be on any causal pathway between EI and CHD, since one can eat more or less energy without necessarily changing fat intake. However, in relation to CHD, the meaning of a change in EI when keeping fat intake constant may be different from considering a change in total EI.
r) Mention that individual diet intake for one week at the start of the study is not really a good index for assessing the energy intake effect / 5 / Long follow up time.
s) Mention of other possible sources of bias (eg. Confounding bias or selection bias) / 5 / For example physical activity, which is not mentioned in this study.
Additional
u) Poisson regression / 10 / Ref: CAL SM11 / streg energy2, dist(exp)
streg energy2 I.fat3, dist(exp) / Note: this approach not essential for full marks
v) Over length / -5 / Note: penalties for going over length in the AA are strict
w) Units / -5
x)Comparing adjusted and unadjusted analyses on slightly different data sets / -3 / Note: this point relates to the approach taken to missing data in the analysis