PANDA

Data Analysis Package

PANDA is a database application designed for clinical trial data processing and statistical table generation, for documentation of the trial and incorporation into statistical reports. This application is based on algorithms and knowledge bases of clinical trial analysis methodology and statistical analysis techniques, which are fully compliant with ICH guidelines.

The main features include:

•  Automatic identification of study populations eligible for primary and secondary efficacy analyses, and safety analysis

•  Automatic selection of observations valid for efficacy and safety analyses

•  Automatic processing of temporal data

•  Automatic data summarization

•  Automatic generation of statistical reports in accordance with ICH guidelines, including tabulations and data listings of baseline data, efficacy data, protocol deviations, compliance checking, adverse events, clinical laboratory safety parameters, among others.

PANDA is a knowledge-based expert system. Its knowledge bases contain representations of the rules governing clinical trial data analysis and the presentation of results. Most rules are formalizations of the ICH-E3 guidelines. Whenever the guidelines were ambiguous or incomplete, the rules were extracted from the specialized literature and adequately documented.

To illustrate the type of content in the knowledge bases, the following figure displays the rules used by PANDA for the identification of the different efficacy populations for which any single study subject may be eligible. PANDA recognizes five efficacy populations (all randomized patients, intention-to-treat, all patients treated, modified intention-to-treat and per protocol) and one safety population.

Rules of that kind were developed for all aspects related to clinical trial data analysis. For example, for the identification of the visits valid for analysis in each study population a complex set of rules had to be defined.

PANDA also embodies algorithms for the analysis of temporal data, which represent the largest part of clinical trial data. Those algorithms are tolerant to missing dates, in particular start and stop dates of concomitant medications, a problem often encountered in clinical trial data. PANDA uses all available information, such as visits dates, to compute the time period when a subject has been exposed to each treatment. The algorithm deals effectively with all kinds of situations, including open time intervals.

PANDA is extremely easy to use. Actually, its interface consists only of a single screen, which is shown in the following figure. The user just needs to select the appropriate study, identify the population to be used in the primary and secondary efficacy analysis, and select the required reports from a list that includes all the tables, tabulations and data listings defined by the ICH E3 guideline.

From then on, PANDA performs all data processing. The application selects the subjects and observations valid for each analysis, queries the COATI database to collect the data needed for report generation, and automatically generates and prints the reports. There is absolutely no user intervention during this whole phase of clinical trial data analysis.

For report generation, PANDA uses Business Objects, a powerful software product for automatic query generation and report preparation that is widely used in data warehousing and data mining.

PANDA creates, in a matter of seconds, complex reports that are error-free and that would take days to produce manually. For example, the following report represents just one of the several tables needed to report adverse events in a statistical report. That table, according to ICH specifications, has to combine information of different nature, such as data on the adverse event itself, the subject's demography, the dose of study medication and the concomitant medication administered during the period of the adverse event.

Many tables in a statistical report require, like in the previous example, the combination of data from different sources. The preparation of such tables imposes the creation of multiple and complex database queries, a process that has to be performed by database programmers.

One of the main advantages of a clinical trials information system like COATI, which is based on a generic model of the clinical trials paradigm, is clearly the elimination of the need to create and validate ad-hoc queries for each type of problem found in each clinical trial.

Another type of statistical tables that are difficult to prepare manually are those containing classification data, as for example the description of concomitant medication. In order to create those tables, one needs to classify each item in the appropriate class, and then count the number of subjects exposed to each item as well as the total exposed to that class. PANDA uses its embedded controlled vocabularies for automatic classification and counting, and presents the results in a compact but perfectly readable table. An example of a table for the description of concomitant medication is shown below.

There are still other tables of higher degree of complexity. All the clinical safety laboratory data, for example, are extremely complex because of the number of data items, the nature of the data and the several difficult analyses required. For example, laboratory data may be in a variety of units, and reference values often vary across study laboratories.

For the analysis of laboratory data, PANDA starts by converting every value into SI units. Next, it searches the reference values adopted by each study laboratory, converts them to SI units and identifies the observations with values outside the reference range. Finally, it creates the requested tables. All this process is entirely automatic and does not require any user intervention. The following example displays one of the tables required by the ICH guidelines for the description of laboratory abnormalities observed during the trial. This particular case displays treatment-emergent laboratory changes.

PANDA does not analyze efficacy data. That procedure is performed by another component of the system, DART (Data Analysis and Reporting Tool). However, PANDA has a role in that process by selecting the data and metadata needed for efficacy analysis, and by exporting those data to DART. Once the data are processed by DART, PANDA imports the results of the statistical analysis and prepared the necessary reports.

PANDA generates automatically all the tables, data tabulations and data listings required by the ICH E3 guidelines. The following list shows the tables prepared by PANDA:

Conduct of the study

Patient accountability

Post-randomization discontinuations

Protocol deviations

Efficacy analysis

Number of patients excluded from efficacy analysis

Evaluable patients

Post-randomization discontinuations (primary efficacy population)

Protocol deviations (primary efficacy population)

Distribution by concomitant illness

Distribution by concomitant medication

Drug dose

Compliance with the study medication

Safety analysis

Dose of test drugs

Duration of exposure

Dose of test drugs: frequency distributions

Adverse events - summary

Analysis of all adverse events

Display of all adverse events

Laboratory values over time

Laboratory measurements: change from baseline

Abnormal laboratory values

Abnormal laboratory values over time

Predefined change values over time

Abnormal vital signs values

Vital signs over time

Primary efficacy population

Distribution by concomitant illness

Distribution by previous illness

Distribution by previous treatment for primary disease

Distribution by treatments stopped at entry

Distribution by previous treatments maintained

Data listings

Discontinued patients

Protocol deviations

Patients and observations excluded from the efficacy analysis

Tabulation of individual baseline data

Tabulation of individual efficacy data

Individual patient's doses

Compliance data

Adverse events by patient: all adverse events

Adverse events by patient: serious adverse events

Adverse events by patient: significant adverse events

Listing of individual laboratory measurements by patient

Individual patient changes

Listing of individual abnormal laboratory measurements

Listing of individual abnormal vital signs measurements