SAS Tip for Categorising Calculated Values

SAS Programming Tip For Categorising Calculated Values

We derived a % result based on 2 values, using the simple formula based on a post and pre assessment:

AVAL=(postf-pref)/pref*100;

For 2 patients, we ran into issues when the following values were used:

subject 001: AVAL = (0.84-0.75)/0.75*100 = 12

subject 002: AVAL = (1.38-1.2)/1.2*100 = 15

Continue Reading

SAS Art Competition 2017

SAS Art Competition 2017 - Open to all

It's that time of year again!

We would like to invite all SAS users to take part in our annual art competition. The winning entry shall receive a £500 shopping token (which can be donated to your chosen charity, if preferred). The entries should be sent to This email address is being protected from spambots. You need JavaScript enabled to view it. by 5pm UK time on Friday 24th November 2017. We would ideally like the artwork to be printable in large format (A0 or 1 metre square approximately) though the file can be sent as either a bitmap or JPEG file in smaller format. Please include the SAS code and graphic in your email, along with any other explanation required to understand and/or interpret the artwork. All entries should incorporate at least one of the colours in the PHASTAR logo (the darker colour is cx73002E and the lighter colour is CXAF95A6).

We are also opening the Christmas competition. The winning entry will receive a £200 shopping token and will also feature on PHASTAR Christmas cards this year.

You can find the results of last year's competition here - SAS ART 2016 results.

Here's an example:

SAS Art Example

Continue Reading

PHASTAR at PSI 2017

PHASTAR staff made multiple contributions to this year's PSI Conference, which was held in London this past May.

Members of our statistical staff presented on Overdispersed Count Data, describing various models available to account for overdispersion, and presented a simulation study comparing their performance when dispersion differs between treatment groups. They concluded that the standard Negative Binomial model is quite robust to violation of the assumption of a single common dispersion parameter. However, in extreme circumstances, fitting a Heterogeneous Negative Binomial model can provide improved standard errors.

Continue Reading

Missing Data at Clinical Trial Level?

Completeness of Clinical Trial Reporting

Professor Sally Hollis, PHASTAR's Head of Statistical Consultancy, was a panellist on an EFSPI/PSI webinar on Data Sharing recently (recording available here: https://youtu.be/V8lsPLck5xI). This is an area which has been rapidly evolving over the last decade. The EMA first outlined steps towards proactive disclosure of data in an article published in 2012 (http://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1001202).  

Following a process of consultation, the EMA published a policy on publication of clinical data in October 2014, the first phase of which came into force in January 2015. This provides access (via the EMA website) to documents relating to submissions, including the individual study reports with redaction of personal data, supporting documents such as the protocol and amendments, CRF and analysis plan, and the clinical overview and summary. In a second phase, the EMA are consulting with stakeholders to find the most appropriate way to make Individual Patient Data (IPD) available, in compliance with privacy and data protection laws.

Continue Reading

Managing Missing Data

Missing data: Management and Prevention

Data managers strive to produce high quality, reliable and intact data for analysis. Integral to this quality standard is to ensure minimal or no missing data. Missing data may have different sources such as equipment failure, missed visits, death or withdrawal of a subject and is usually dealt with during the analysis by defined handling strategies. Data which are available at the investigational site, but have not been collected and are missing from the eCRF through error or omission can be avoided by good data handling procedures.

The impact of missing data can be many fold from delay in timelines, additional costs and resources associated with retrieving and reconciling the data, and of course, adversely affecting the interpretation of study results through the introduction of bias. Many data items are dependent upon or form dependencies on other data items, therefore the unavailability of a single item of data may affect the integrity of data points elsewhere in the database.

The optimal approach to dealing with missing data is one of prevention. Trial design has a role to play and consideration should be given to practicalities, such as the impact on site and subject, in an effort to avoid missing data due to confusion or errors in study conduct. Effective and efficient data capture processes are essential. Good eCRF (or paper CRF) design with a logical data flow which mimics the sequence of procedures in the clinic and facilitates efficient data collection is important. Skip logic is a feature that changes what question or page a respondent sees based on how they answer another question, thus guiding the user through the eCRF and avoiding data entry into variables that should remain blank. Clear on-screen data entry instructions and readily available eCRF completion guidelines are essential. User Acceptance Testing (UAT) during the database design stage is key to this and developers should ensure that a wide range of user-types test the design to consider ease of use. It goes without saying that user training and support are essential. Along with periodic refresher training demonstrations, vignette style videos are well received.

Continue Reading