代写ST332/ST409 Medical Statistics 2023-24: Practical 1帮做R程序

ST332/ST409 Medical Statistics 2023-24: Practical 1

Clinical Efficacy of Orlistat Therapy in Overweight and Obese Patients With Insulin-Treated Type 2

The aim of this study was to assess the effect of orlistat, a gastrointestinal lipase inhibitor, on weight loss, glycaemic control, and cardiovascular risk factors in overweight or obese    insulin-treated type 2 diabetic patients. The original trial paper can be found in Diabetes Care 2002 Jun; 25(6): 1033-1041.https://doi.org/10.2337/diacare.25.6.1033& there is a pdf on Moodle ST332 page.

This study was a 1-year multicentre, randomised, double-blind, placebo-controlled trial of orlistat (120 mg three times a day) or placebo combined with a reduced-calorie diet in overweight or obese adults (BMI 28–40 kg/m2) with type 2 diabetes treated with insulin alone or combined with oral agents, but with suboptimal metabolic control (HbA1c 7.5– 12.0%). The only stratification factor was HbA1c<=10%/HbA1c>10%.

Outcome measurements included changes in body weight, glycaemic control, blood pressure, and gastrointestinal adverse events (AEs).

Variables

orlistat:  Diet+Placebo=0 Diet+Orlistat=1

Age: age in years

bmi: (Body Mass Index, weight[Kg]/(height*height [m], 25-

30=overweight, 30+=obese)

female: Male=0 Female=1

caucasian: ther=0 Caucasian=1

hba1c: HbA1c<=10%=0 HbA1c>10%=1

weight.change: change in weight from baseline (Kg) – negative

values indicate a loss in weight (compared to baseline)

gastro.ae: gastrointestinal adverse events, no=0, yes=1

From the ST332 Moodle page you should download the R dataset ot2d.RData –  note that this is a synthetic version of the trial data so your results will not be exactly the same as those reported in the paper but should be similar.

1.   In order to assess whether randomisation has worked look at the baseline variables (age, bmi, female, hba1c and caucasian) in each of the two treatment groups. Does randomisation appear to have worked? [Hint: use the table, prop.table, mean and sd functions in R/RStudio)

2.   For weight loss: (i) calculate the mean (and the standard deviation) in each of thetwo treatment groups separately, (ii) produce a boxplot for weight loss for each of the two treatment groups, (iii) perform. a 2-sample t-test to compare the weight loss in the two treatment groups. What evidence is there that orlistat is better than placebo?  [Hint: use the mean, sd, boxplot and t.test functions in R/RStudio]

3.   For gastrointestinal adverse events: (i) estimate the probability of having a

gastrointestinal adverse event in each of the two treatment groups, and (ii) compare them using a Chi-squared test. What evidence is there that the probability of having  a gastrointestinal event is greater if a patient takes orlistat compared to placebo?

[Hint: use the table, prop.table and chisq.test functions in R/RStudio] What is surprising as regards the Diet+Placebo group?

4.   Repeat the analyses in Questions 2. and 3. but only for women in the trial. What

evidence is there that effect of orilstat is different in women compared to the overall trial population? [Hint:

ot2d$orlistat=="Diet+Orlistat"&ot2d$female=="Female" will identify women who take orlistat]

5.   The trial used HbA1c<=10%/HbA1c>10% as a stratification factor, and therefore to

attain maximal statistical power this should be adjusted for in any analyses. Using a linear model repeat the unadjusted analysis of weight change for Question 2 and check that you get the same answers. Then extend the model to also include HbA1c – what is the effect of the adjustment on the treatment effect and the level of uncertainty? Further extend the model to assess whether the effect or orlistat differs between males and females [Hint: you will need to fit an appropriate interaction model].

6.   Using a generalised linear model (GLM) conduct an unadjusted analysis of

gastrointestinal AEs as in Question 3 – what is the Odds Ratio (OR) [and 95% CI] for

the risk of having a gastrointestinal AE in the Diet+Orlistat group compared to the

Diet+Placebo group? [Hint: you can use exp(cbind(coef(x), confint(x))) where x is a GLM model fit object to calculate the OR and 95% CI].  Extend the model to also include HbA1c – what is the effect of the adjustment on the OR and 95% CI?

7.   If a parent or grandparent was overweight or obese and had sub-optimally

controlled type 2 diabetes write a one sentence summary of the key trial results for them.

Questions 8 & 9 require you to have covered sample size calculations in the next lecture … attempt them afterwards.

8.   A clinician asks you to help design a future follow-on trial (for a similar patient

population), but this time with a Minimum Clinically Significant Difference for the difference in weight loss (from baseline) between orlistat and placebo of 1Kg – calculate how many patients would be needed for this new trial (assuming a 5% significance level and 80% power) [Hint: use the power.t.test function in

R/RStudio].

9.   Worried about the increase in adverse events, another colleague suggests that even a 10% increase over placebo would be concerning – how many patients would be

required to detect such a difference in this patient population (assuming a 5%

significance level and 80% power)? [Hint: use the power.prop.test function in R/RStudio]