Read time 5 min

Once the data collection is completed, the data have been cleaned and the database is locked, the analysis phase can start. The outcome of the study is analysed and the result compared with the original question.

It is a good idea to consult a statistician already during the planning phase to enable the production of correctly analysed data and publication of the result. Early preparation is crucial for producing a good study design, the correct number of study participants and the right evaluation measures to show what it is you want to achieve.

The ICH and GCP guidelines E3 (pdf, 471 kB)external link, opens in new window and E9 (pdf, 284 kB)external link, opens in new window include guidelines for statistical principles for clinical studies. Therapy studies are expected to follow these guidelines, and it is recommended that all clinical studies follow the guidelines as much as possible.

Statistical analysis plan

A statistical analysis plan (SAP) that includes a detailed analysis description of all variables shall be designed and ready before the database is locked for a blind study, and before the study begins for an open study. The plan shall have a version number and be signed with date and time by the SAP author, the clinical investigator and/or the sponsor. The plan shall be saved for any future scrutiny.

Analysis population

In a clinical study, data shall be analysed using several different analysis populations.

The Intention-to-Treat (ITT) population shall include all the randomised/included individuals in the study. All effectiveness analyses shall be carried out on the ITT population. Sometimes, you may have to use a modified ITT population, a “full analysis set” (FAS) population, where individuals who do not have any follow-up data, for example, are excluded. In the ITT analyses, the study participants are analysed according to the therapy they have been randomised for, irrespective of what therapy they have then actually received.

Per protocol (PP) population includes all the individuals who are randomised in the study, and who have not had any serious protocol discrepancies during the study period. The PP population is set at a clean file meeting before the database is locked. All or a selection of the effectiveness analyses is then also carried out of the PP population to show how robust the main results of the study are. In the PP analyses, the study participants are analysed according to the therapy the study participants have actually received.

The safety population includes all individuals who have received at least one dose/treatment in the study. In the safety analyses too, the study participants are analysed according to the therapy the study participants have actually received.

Replacement (imputation) of missing values

It is important to specify in advance how missing values, “missing data”, shall be handled in the analyses. Some examples of imputation methods are:

  • simple imputation, such as mean imputation, last observation carried forward (LOCF), worst observation carried forward (WOCF)
  • stocastic imputation, or random imputation of a value based on a regression model that includes the patients’ background data
  • multiple imputation, that is a stocastic imputation carried out on at least five sampled studies with stocastic imputation, where the variability of the results is adjusted by pooling the results of each individual study

Statistical analysis

The statistical part of the protocol and SAP should specify the analyses to be tested and/or the therapy effects to be calculated in order to fulfil the primary goal of the study. The primary analysis of the primary variable should be separated clearly from the supporting analyses of the other variables. Effect measurement, variability measurement and p value shall be presented.

Examples for effect measurements are:

  • mean
  • median
  • least square (LS) mean
  • proportion
  • risk ratio (RR)
  • odds ratio (OR)
  • hazard ratio (HR)

Examples for variability measurements are:

  • confidence interval
  • standard deviation
  • standard error
  • inter-quartile range (IQR)
  • minimum and maximum

In the event of multiplicity, the adjustment of the type 1 error shall be defined in the SAP. Multiplicity may arise in studies with several primary outcome variables, multiple comparisons of therapies, repeated tests over time and/or during preliminary/interim analyses, for example.

Adjustment analysis

It is important to identify and define in advance any confounders and other covariates for which the main analyses must be adjusted. A confounder is a variable that affects both the outcome and the prediction variables. You can identify confounders and other adjustment variables in a structured way by working with directed acyclic graphs (DAGs).

Subsidiary group analysis and interactions

The primary outcome variable may be related to other variables in the study, apart from the study therapy. There may, for example, be relationships with age, gender, BMI, and so on. In randomised clinical studies, it is important to use subsidiary group variables that are measured at the start of the study, that is before the therapy effect has started to work. Subsidiary group analyses are usually of an explorative nature, and should therefore be defined as such in a clear way.

Sensitivity analysis

In order to assess how stable the study results are, it is usually recommended to carry out sensitivity analysis, at least for the primary end point of the study. Examples of sensitivity analysis are:

  • PP population analysis
  • complete case analysis, which includes all individuals who have data at all visits in the study
  • one or several analyses of data where missing values are replaced with the help of various imputation methods

Interim analysis

Interim analysis is analysis that is carried out during while the study is in progress, and is aimed at providing a recommendation whether the study should continue to be terminated early. As the performance of an interim analysis impacts on the interpretation of the study results, it is important that any interim analysis is planned carefully already in the study protocol. This type of analysis is usually carried out by the study’s Data Safety and Monitoring Board (DSMB), which is an independent monitoring group. Any de-blinding shall be kept within the independent group, and all personnel working with the study shall remain blind to the result of such analysis. Interim analysis that is carried out in an unplanned manner should be avoided, as it can influence the results of the clinical study and weaken the trust in the conclusions drawn.

Analysis database and analysis program

The final analysis database and analysis program should be archived for possible future scrutiny and control.

If study data is to be submitted to the medicine authorities in USA (FDA) or Japan (PMDA), the study database shall be designed according to the CDISC (Clinical Data Interchange Standards Consortiumexternal link, opens in new window) standardised format.

Evaluating the result

Once the analysis has been concluded, the results shall be evaluated and a number of questions asked, such as:

  • Has the original question been answered?
  • What conclusions can be drawn, and what are the consequences?
  • Have any new questions arisen during the process?
  • What shortcomings were there in the methods used?
  • Did you get an answer to something you had not asked about?

Statistics Sweden has a statistics guide (in Swedish)external link on its website.

Writing a study report

For clinical medicine studies, a detailed study report is required for each individual study. The study report shall be written according to the prescribed regulatory guidelines. The guidelines are available on EMA’s website at “Structure and Content of Clinical Study Reports” (pdf, 370 kB)external link, opens in new window.

Date created: 2017-11-14
Last published: 2018-05-24


Office for Clinical Studies

Östra Hamngatan 26
SE-411 09 Göteborg


Phone: +46 (0)31 757 41 75

E-mail: kliniskastudier@vr.se


Would you like to subscribe to news, good examples or calendar events from clinicalstudies.se?

Clinical Studies Sweden - a collaboration between The Swedish Research Council and the Swedish Association of Local Authorities and Regions.