Importance of statistics in randomised controlled trials (RCT)

Quantitative studies are those that rely on measures that can be represented by discrete numbers, such as age, weight or body temperature. Epidemiological studies too can be quantitative in nature. A quantitative epidemiological study can be broadly classified as ‘observational’ or ‘experimental’. It depends upon the extent of intervention by the researcher in the subject’s exposure or actions. Observational studies are further classified as ‘descriptive’ and ‘analytical’. The flow chart below gives a brief idea of the classification of epidemiological studies. In this article, common statistical tools and techniques used to study the data gained from randomised controlled trials (RCT) or clinical trials is studied.

Source: Pearce (2012)

Types of randomised control trials (RCT)

The ability of the study design to prove the evidence shows that RCTs’ are at the apex of the research pyramid. The randomisation of data takes care of known and unknown confounders and bias (Moher et al., 2010). RCTs’ are rigorously conducted studies which have a set of schools in terms of statistical analysis. A statistical analysis of an RCT can be conducted as ‘Intention to treat’ (ITT) or ‘Per Protocol’ (PP). ITT is a strategy wherein data on all the participants who were enrolled in the trial after random allocation must be analysed regardless of their status (cured, death, drop out, loss to follow-up) at the end of the trail. PP is a strategy wherein the data form non-compliant (dropouts, loss to follow-up) participants are not analysed (Wassertheil-Smoller and Kim, 2010).

Source: Parikh et al. (2007)

The figure above shows the process of a clinical trial conducted on testing the effect of an anti-fungal medicine Fluconazole on underweight infants. Initially, 700 participants were randomly allocated to either intervention (Fluconazole) or control (placebo). However as the trial progressed, several subjects were lost, either in not receiving treatment or not following-up on the treatment. At the end of the trial, if the results are analysed using ITT strategy,  data on all the 350 randomly allocated participants will be considered, even though they were non-compliant.

Common concepts to be applied for analysing RCT data

During the analysis of clinical trial data, there are several rules or concepts that are required to be considered, some of which have been described below (Moher et al., 2010):

1. Understand the type of data: The data collected for RCT or in case of any epidemiological study can be broadly classified into ‘categorical’ (age groups, gender) and ‘continuous’ (height, weight). It is important to understand the data category because it helps understand the analysis method.
2. Understand the dependent (outcome variable) and independent variable (exposure variable): Dependent variable is the outcome of the intervention applied. Independent variable can be anything from a risk factor to a disease (smoking), a condition (hyperglycaemia), or a disease (malaria) which can affect the outcome.
3. Understand the relation between the dependent and independent variable: There are instances when a dependent variable is categorical and independent is continuous, the type of statistical tests to be applied varies due to the type and relationship between these two players. The table below summarises the statistical tests to be applied to various permutations and combinations of dependent and independent variables.

Source: (Nayak and Hazra, 2011; Parab and Bhalerao, 2010)

4. Understand the statistical software: Statistical software packages are available in abundance in the market which can analyse the data. Also, each of them has certain unique characteristics which help in their selection, based on the type of study and the type of analysis to be conducted. The table below lists the best compatible software based on the study type.

Common pitfalls in statistical analysis

Along with the advantages of applying statistical tools, it is also important to avoid mistakes in choosing a test, conducting a statistical test or even during the interpretation of the results. There are several pitfalls to be avoided while applying statistical tests on clinical trial data, as discussed below (Charan and Saxena, 2012):

1. Managing missing data: Missing data on a large number of trial patients leads to skewing of results on either side. Care must be taken while dealing with a data set which has a large number of missing entries. The integrity of the clinical trial rests on the completeness of the data.
2. Applying the wrong statistical test: Every statistical test has its own requirements, when the data fulfils the requirement of the test then only the results from the tests can be trustworthy. In cases where data doesn’t satisfy the condition of normality, non-parametric tests should be a test of choice.
3. Reporting of test results: A test result for the continuous variable must be reported with mean or median, p-value, and 95% confidence interval. While the test results for the categorical variable must be reported as a percentage with sample size.
4. Ad hoc reporting of subgroup analysis: CONSORT clearly states that the trial protocol should clearly mention the number of subgroup analysis to be performed and post hoc subgroup analysis should be avoided

Overall, the statistical software to be used in quantitative epidemiological studies will be based on the aim of the study, the type of study and the type of data.

References

• Charan, J., Saxena, D., 2012. Suggested Statistical Reporting Guidelines for Clinical Trials Data. Indian J. Psychol. Med. 34, 25. doi:10.4103/0253-7176.96152.
• Dembe, A.E., Partridge, J.S., Geist, L.C., 2011. Statistical software applications used in health services research: analysis of published studies in the U.S. BMC Health Serv. Res. 11, 252. doi:10.1186/1472-6963-11-252.
• Gupta, S.K., 2011. Intention-to-treat concept: A review. Perspect. Clin. Res. 2, 109–112. doi:10.4103/2229-3485.83221.
• Moher, D., Hopewell, S., Schulz, K.F., Montori, V., Gøtzsche, P.C., Devereaux, P.J., Elbourne, D., Egger, M., Altman, D.G., 2010. CONSORT 2010 Explanation and Elaboration: updated guidelines for reporting parallel group randomised trials. J. Clin. Epidemiol. 63, e1–e37. doi:10.1016/j.jclinepi.2010.03.004.
• Nayak, B.K., Hazra, A., 2011. How to choose the right statistical test? Indian J. Ophthalmol. 59, 85–86. doi:10.4103/0301-4738.77005.
• Parab, S., Bhalerao, S., 2010. Choosing statistical test. Int. J. Ayurveda Res. 1, 187–191. doi:10.4103/0974-7788.72494.
• Parikh, T.B., Nanavati, R.N., Patankar, R.V., Pn, S.R., Bisure, K., Udani, R.H., Mehta, P., 2007. Fluconazole Prophylaxis against Fungal Colonization and Invasive Fungal Infection in Very Low Birth Weight Infants. Indian Pediatr. 44, 830–837. doi:10.1.1.554.3205.
• Pearce, N., 2012. Classification of epidemiological study designs. Int. J. Epidemiol. 41, 393–397. doi:10.1093/ije/dys049.
• Schulz, K.F., 2010. CONSORT 2010 Statement: Updated Guidelines for Reporting Parallel Group Randomized Trials. Ann. Intern. Med. 152, 726. doi:10.7326/0003-4819-152-11-201006010-00232.
• Wassertheil-Smoller, S., Kim, M.Y., 2010. Statistical Analysis of Clinical Trials. Semin. Nucl. Med. 40, 357–363. doi:10.1053/j.semnuclmed.2010.04.001.