Clinical Data Management Tips

Dear Reader,

This article is about black box testing which I use a lot within clinical data management. It helps me to judge if system, people and procedures are able to assure quality clinical study data.
Therefore I do not exactly need to know how things are done. Instead I focus on what is done. Output versus input. Simply said, I check if what goes IN, is coming OUT as expected.
I use it within clinical studies to check accurate database preparation as well as for testing programmed edit checks. I use it for quality control activities to check accurate clinical study data handling. And for test scripts, carried out for computerized system validation.
Almost all clinical data management activities can be checked with the IN = OUT, unless… principle. 

Best regards,

Maritza

1. Working harder. Spending more time in the laboratory, in the office, at your desk. If you manage to work more, as a result, you will only get more of the same kind on your to do list. Working more is fine for a short period with a clear goal. With a week max! Any longer works like an energy-drain. Slowly leaking more and more energy, fun and possibilities out of your role.
2. Hoping that the future will contain that time for you. Well, the future hasn’t, if you do not start yourself. If you do not change, the future does neither.

1. Reserve time to sit down, without distraction to invest in your idea. Even though your regular tasks are not yet finished and urgent requests were asked, guard the time scheduled for your creative mind. If you long to make a difference, this time is equally important!

Start to reserve one to 2 hours a week and just take that time. Even though you don’t know how to begin implementing your idea, take that time to start. You need to learn (again) how to put ideas in to practice. You’ll soon notice which steps are needed next.(Phone’s still and out of sight, door shut.)

2. Create space for creativity in your head. David Allen’s Getting Things Done is a very good book to start creating space in your mind again.

What can count on your enthusiasm, your interest, within clinical research? The sophisticated technology, the ambitious people challenging themselves, the added value to patient care & health? Where do you walk that extra mile for? For what clinical results and ideas do you step up?
Once in a while I encounter these questions again. What drives me to write these articles, what attracts me in clinical research? Lately because I realized I’m not deeply within core clinical data management anymore. But I still do care for people with all kinds of different roles in clinical research. How all these various specialists together get medicines and devices approved for release. From the Principal Investigator and Research Nurse at the study site, the Monitor, Project Manager, Data Manager and Statistician to the Medical Writer of the Clinical Study Report. And all other (management) functions in between. Not forgetting the more staff like roles QA and IT people conduct.
And I do think you still experience challenges comparable to mine. Ever wanted extra space to grasp an indication, a diagnosis and its implications for people’s health? Ever longed to step up for a piece of software that could boost the lay-out and receipt of your reports? Ever wanted to acquire a SAS BASE license to use within your group? When I am enthusiastic of an idea or a capability, I easily walk extra miles to get this idea or chance heard and given feedback up on. This weeks article is therefore about how to step up for your clinical results and ideas. If you want to get heard too, this article is for you.

How to step up for your clinical results and ideas – 2 strategic steps

How to present your clinical study results and your clinical research ideas at meetings, during presentations, at tele- and video conferences etc. so that your message is received? Simply heard and given feedback up on?

Well, not by:

1. The most fancy visuals. Although gadgets, easy effort and a sense of humour can help to brake any initial ice.
2. Adjusting yourself to your audience by suddenly wearing a complete suit!, for example. Changing the look you normally spread professionally, doesn’t contribute to your story. On the contrary, it distracts and weakens your story. Even if your audience is completely new to you. So, if you normally can wear jeans and a simple jacket; take care but keep that look.
3. Changing your moves. For example your normal way of talking. Trying to speak slow whereas you normally enthusiastically speak with heights and troughs. Or trying to move your hands less, whereas your hands naturally tend to help telling your story.

Two strategic presentation preparation steps:
 

1. What exactly do you want to achieve with your presentation? Do you want higher salary scales for your department? In line with your gained additional tasks and / or responsibilities? Do you want to acquire a second, lighter EDC system for your post marketing studies? Do you want to acquire a SAS license? To create raw data listings with for your Clients? Do you want to lessen the effort spend to complete your trial’s CRFs? What do you want to gain? Where do you see an opportunity for your organization or the clinical trial results? How would that make life easier?

In fact, the one big heart felt item that GOT you step up?

2. What does your audience need to understand your message? Do they need facts and costs? Impressions? Organization impacts, FTE’s? What do they need in order to be able to decide up on your subject? Who is your audience? Is that the correct audience? Can they decide themselves or can they help you further? Should you change audience? Directly contacting these managers yourself? Or in the complete picture; resonates your idea with the organization’s core business?

Imagine being part of your audience. Take in to account people’s roles within the audience you are presenting to.

Any idea what such presentations would end in?
Feedback and support to get (part of) your idea implemented. Or a very reasonable reason to park it or skip it (for the moment). Very important, allowance to participate in the discussion of your idea. And on term, more opportunities to use your dedicated strengths.

You can check if I made a good presentation start…., if you hear me through the “Significant presence program; 8 strategic steps to succeed in sharing your clinical results and ideas!”

Wishing you space to donate YOUR research contributions,

best regards,

Maritza

© 2011, Maritza Witteveen, ProCDM

You’re welcome to re-publish this newsletter, but please add the following text to it. This is an article from Maritza Witteveen. Helping enthusiastic Clinical Research Professionals to contribute to clinical results. You can receive more articles and the free e-book ‘Five strategies to get reliable, quality clinical data’ by subscribing to my ezine via http://www.procdm.nl/pages/knowledgebase.asp.

Last Friday I red an article about a new book of John Tierney en Roy Baumeister: Willpower. Rediscovering the Greatest Human Strength. The article struck me because it explained why my texts so often contain words like ‘structure’, ‘format’, ‘method’ and ‘program’.
Schedules and templates to work with give me room to contribute my time to the extra ordinary. E.g. to listen carefully to Clients’ needs, to find and improve solutions, to answer complex questions. In order words; to focus up on out of ranges, strange information, and phrasing.
But the article took the benefit of structures a step further by adding that structuring the ordinary saves energy. Energy, willpower you then can use to achieve your goals. Because with structures, you do not need to consume time and energy to the WHAT, WHEN, WHY, WHO and HOW of routine steps. Taking decisions, defining when it needs to be met. No, that’s already there. The decisions left are those that specifically benefit to the particular project under subject. Content decisions. Therefore this weeks article about structuring soft skills for clinical results.

 

One of the three main tasks of data management is to translate individual subject data to logically grouped datasets ready for analysis. Study data captured in a structured format with which the statistician can work. But with what datasets can the statistician work?
In fact, with everything. Because he or she is capable of transforming your datasets to suiting datasets with the statistical software. So the question could better be, with what datasets can the statistician comfortably work? Without re-structuring the data delivered?

Well, first it is handy to know a bit more about the products statisticians deliver.

1. Tables with descriptive statistics describing the subject group under study. Overall descriptive stats for all subjects together or descriptive stats per treatment group or per gender.

N=208
Gender          Male 84 ( 40,4%)                 Female 124 (59,6%)
Age (yrs)        Mean = 56,7                         Min = 34   -    Max = 82
Weight (kg)   Mean = 78,8                         Min = 51   -    Max = 111

Tables for safety outcomes. Numbers and percentages of adverse events that occurred. Overall and per treatment group.

Adverse events                   Medication A      Medication B
                                                    (n=1205)               (n=1200)
                                         No. (%) of patients   No. (%) of patients
Gastrointestinal disorders  101 (8,4%)                            113 (9,4%)
  Diarrhea                              67 (5,6%)                               66 (5,5%)
  Nausea                               61 (5,1%)                               57 (4.8%)
Muscoloskeletal and connective
tissue disorders                   98 (8,1%)                               89 (7,4%)
  Pain in extremity                45 (3,7%)                               59 (4,9%)
  Back pain                            72 (6,0%)                               62 (5,2%)

2. Graphs to visually compare the different intervention groups under study. E.g. survival rates, pharmacokinetics.

3. Statistical tests to compare the efficacy objectives between the different intervention groups. On which the conclusion of your clinical study report will be based.

“Subjects receiving the new medicine were significantly
more likely to respond well up on overall quality of life, than were those who received the placebo(P < 0,05), whereas those walking within 24 hours after surgery, or weight loss were no more likely to respond well than those without these features.”

4. And last but not least, raw data listings if not already created by data management.
(The advantage of creating raw data listings for a study is that you get to know the individual study data. You are busy with all individual data records, instead of grouping them into a table, graph or analysis. It helps to get to know the individual drop-outs, the outliers and the missing measurements.)

Subject number  Visit date       Diastolic blood pressure  Heart rate
                                                           (mm Hg)            (bpm)
1209                        13AUG2011        127                            78
1210                        15AUG2011        116                              89
1301                        16JUN2011         104                             91
         

This about the products statisticians deliver for a clinical study report. Secondly some examples of datasets and why chosen as such:

1. A demography dataset, DEMO, is delivered with all demography data for all subjects, like gender, date of birth, but also subject number and date of screening. Only this demography dataset is needed to program a descriptive stats table for all subjects.

SUBJID         DSCREEN   GENDER      DBIR
1209              12JUN2011   1               17OCT1945
1210              13JUN2011   2               10FEB1961
1301               07JUL2011   1               04DEC1954

In- and exclusion criteria can be a separate dataset. Because these are only listed and checked for deviations.

2. Datasets contain subject numbers and most of them also have visit dates. These so-called key data fields, are used to combine data from different datasets. E.g. a dataset revealing the actual treatments merged with the demography dataset. Using the subject number, both datasets can be combined. And a descriptive stats table of the subjects per treatment group can be programmed.

SUBJID         GROUP         TRTLABEL
1209               A                     New medicine
1210               A                     New medicine
1301               B                     Placebo

With exception of the key data (subject number, visit number), CRF data should exist in one dataset only. Either in this or that, but not in two or more datasets.

3. Another example, blood – and urine laboratory assessments for all visits combined in one dataset. To check for laboratory result shifts across visits.

SUBJID   DVISIT             LABP             LABR   UNIT          OUT    CS
1210        13JUN2011   ASAT              68           U/L              2          2
1210        20JUN2011   ASAT              123        U/L             1          1
1210        08AUG2011   ASAT              72          U/L             2          2
1210        15AUG2011   ASAT               52          U/L             2          2
1210         13JUN2011   Creatinine     69   umol/L             2          2

All measurements collected in one visit are not necessarily present in one dataset. On the contrary, it is more logical to have different measurements in separate datasets. Maybe a measurements dataset for small repeating measurements.

4. Datasets that needed normalization, like often is more convenient for medical history, in- and exclusion criteria and laboratory datasets, can not be combined with non-normalized data in one dataset. Normalized datasets have additional key fields next to subject number and visit number. E.g Criteria number for an in- and exclusion criteria dataset. Or a specimen (blood/urine) field and a laboratory test field for a laboratory dataset.

SUBJID   DVISIT           CRITNO    CRIT                              INEX
1210        13JUN2011  4                   BMI < 25                     Yes
1210        13JUN2011  4                   BMI < 25                      Yes
1210        13JUN2011  5                   Is the subject pregnant?  No

Thus the single outcome of the one-time measured pregnancy test at screening is often added to the demography dataset instead of added to the in- and exclusion criteria dataset.

5. For identification and search reasons, adverse event and concomitant medication datasets contain adverse event numbers respectively concomitant medication numbers.

SUBJID  CONM No.    Medication                 Reason given   AE No.
1301         23                  Atenolol                       Prophylaxis
1301         24                  Prednisone                  Adverse event    3
1301         25                  Acetylsalicylic acid  Adverse event   12
         

Do you get an idea of how to structure your CRF data in logically grouped datasets?
In practice, get the blank CRF and sit down with the statistician or statistical programmer to logically group all CRF data in datasets. The total number of datasets for a regular clinical study…. is around 20 to 30 different datasets. Estimated time to draw CRF data to grouped datasets; 30 minutes. And you will discover with what structured format the statistician comfortable works.

Kind regards, Maritza

© 2011, Maritza Witteveen, ProCDM
 

You’re welcome to re-publish this newsletter if you add the following text to it. This is an article from Maritza Witteveen of ProCDM. Data management for clinical research. Receive tips and the free e-book ‘Five strategies to get reliable, quality clinical data’ by subscribing via http://www.procdm.nl/pages/knowledgebase.asp.
 

How to write queries unambiguously expressing what is asked for?
Using short, polite sentences?
Objectively explaining the underlying inconsistency?
 

First of all my general guidelines.

1. My preference is to use no more capitals then needed. Capitals in the middle of a query text, e.g. for CRF fields or for tick box options, could distract from getting the actual question asked. E.g. compare the same query texts, with and without extra capitals.
   Please verify stop date. (Ensure that stop date is after or at start date and that stop date is not a future date.)
   Please verify Stop date. (Ensure that Stop date is after or at Start date AND that Stop date is not a future date.)
    
2. Referring to CRF fields as they are shown on the CRF. To easily find the involved field(s).
    
3. I prefer to leave any ‘the’ before a CRF field referral out of the query text. For more to-the-point query texts. E.g. compare the same query texts, with and without ‘the’ before data fields.
   Please verify stop date. (Ensure that stop date is after or at start date and that stop date is not a future date.)
   Please verify the stop date. (Ensure that the stop date is after or at the start date and that the stop date is not a future date.)
    
4. Consistency in phrasing a query text can help to quickly write query texts or pre-program query texts in a structured, familiar way. That’s the thought behind the following 6 template sentences for query texts. Which you can use to help you write or program your queries.

The six ‘template’ sentences for query texts:

1. Please provide…

For asking the study site people to provide required data from patient care recordings. Examples:
Please provide date of visit.
Please provide date of blood specimen collection.
Please provide platelet count.
Please provide % plasma cells bone marrow aspirate.
Please provide calcium result.

2. Please complete…
   For asking the study site people to complete required data as required by the study CRF design. (Not necessarily required for patient care). Examples:
   Please complete centre number.
   Please complete subject number.
   Other frequency is specified, please complete frequency drop-down list accordingly.
    
3. Please verify…

For asking the study site people to check date and time fields fulfilling expected timelines. Or for asking the study site people to check field formats. Examples:
Please verify start date. (Ensure that start date is before date of visit.)
Please verify stop date. (Ensure that stop date is after or at start date and that stop date is not a future date.)
Please verify date of blood specimen collection. (Ensure that date of blood specimen collection is before or equal to date of visit and after date of previous visit.)
Please verify date last pregnancy test performed.
Please verify date of informed consent. (Ensure date of informed consent is equal to date of screening or prior to date of screening.)
Please verify date as DDMMYYY.

4. …., please correct.

For asking the study site people to correct a data recording inconsistent with another data recording. Example:
Visit number should be greater than 2, please correct. 

5. …., please tick…

For asking the study site people to complete required tick boxes. Examples:
Gender, please tick male or female.
Pregnancy test result, please tick negative or positive.
Any new adverse events or changes in adverse events since the previous visit, please tick yes or no.
Laboratory assessment performed since the previous visit, please tick yes or no.
LDH, please tick normal, abnormal or not done.

6. Please specify…

For asking the study site people to specify the previous data recording. Examples:
Please specify other dose.
Please specify other frequency.
Please specify other method used.
Please specify other indication for treatment. 

Finally, for query texts popping up during CRF data recording, it could be helpful to put location information in it. Like:
Page 12: Please verify start date. (Ensure that start date is after or at start date on page 11.)

Good luck finding your way to structure query texts,
kind regards, Maritza

© 2011, Maritza Witteveen, ProCDM
 

You’re welcome to re-publish this newsletter if you add the following text to it. This is an article from Maritza Witteveen of ProCDM. Data management for clinical research. Receive tips and the free e-book ‘Five strategies to get reliable, quality clinical data’ by subscribing via http://www.procdm.nl/pages/knowledgebase.asp.

Which documents are critical for showing that your clinical data management system (CDMS) is in a validated state? Which documents are essential to be able to generate documented evidence?

5 important documents surrounding the CDMS:

1. The User Requirements Specification (URS). Needed to be written even before a CDMS is chosen! In fact, this document is written to choose a suitable CDMS with. You can categorize your requirements and wishes as (1) Must have, (2) Should have, (3) Would have, (4) Could have (MosCoW classification system), to indicate the necessity of each requirement. With this URS you then (1) ask potentially suitable CDMS Suppliers to react up on your requirements list as capable of yes or no, (2) accordingly invite Suppliers for a system demonstration and (3) experience potential systems yourself as suitable yes or no.

2. The Computerized System Validation Policy. An organization wide policy with regards to the company’s computerized system validation approach. This policy document should at least be reviewed and approved for by the management of the company. Preferably also initiated by the management team. Because the way computerized system validation is approached, could be critical for the company’s core business. E.g. with regards to a CDMS; critical for acceptation of the results from clinical trials by authorities. What’s the risk if…
A company wide policy too, because more computerized systems, next to a CDMS, could be subject to validation. E.g. systems used in the lab or used for product manufacturing.

3. The Installation Qualification (IQ) Protocol. The plan to check correct CDMS installation with. Often a document supplied by the CDMS Supplier which is used to check each installation step with for correct installation. It should be signed for and dated by at least the performer when installation finished.

4. The Operational Qualification (OQ) Protocol. In fact, the user acceptance testing part. To test correct CDMS performance as the Supplier intended the performance to be. As such, often supplied by the Supplier.
User acceptance tests could be used also for new employees as an user introductory training to get familiar with the CDMS. Because a user acceptance test touches all user features of the system.

5. The Performance Qualification (PQ) Protocol. The plan to check the CDMS system with as it will actually be used in your organization. For this PQ plan, you need to get back to the User Requirements Specification you’ve made before acquiring a CDMS! It lists what you needed a CDMS to be capable of, for your specific organization, before you became familiar with the acquired CDMS features.
This PQ plan; you write yourself. Based on the URS and your clinical data management SOPs for collecting, structuring and verifying study data. The main part will be the actual organization specific PQ scripts, test scripts. Which need execution for documented evidence that the system consistently does what you’ve intended – and expect it to do.

The beginning of the computerized system validation documentation generation starts directly after raising the idea; is it time to acquire our own CDMS? Writing the User Requirements Specification, listing your wishes and requirements for a suitable CDMS. To finally consistently proof that your CDMS does what it is expected to do.

If you want to view an example SOP for the computerized system validation process of a CDMS, please visit http://www.dataentryexport.nl/pages/DEE%20SOPs.html and click on the link of STEP 2.

© 2011, Maritza Witteveen, ProCDM

You’re welcome to re-publish this newsletter if you add the following text to it. This is an article from Maritza Witteveen of ProCDM. Usable data management for clinical research. Receive tips and the free e-book ‘Five strategies to get reliable, quality clinical data’ by subscribing via http://www.procdm.nl/pages/knowledgebase.asp.

Which signs provide me confidence that study data collection, data verification and data cleaning are running as they should? These are simple signs. Signs I get from the data itself, signs from the people recording and delivering the data, and  metrics (signs) from the clinical data management system. Most information is already measured by the clinical data management system and can be viewed through system (status) reports.
One sign although, I experience in study meetings. However, this ‘sign’ is also documented; written down in meeting notes, e-mails and telephone reports.

What are these numbers, information, calculations that indicate data collection is doing fine?

1. The fact that subjects are continuously enrolled per study site. Amongst others, this is an indication that the CRF used to collect data is clear and user friendly. People are not holding back to include subjects because of difficulties they have with the CRF, logistics and/or the queries to expect. The study is progressing and people are all working towards completion.
2. The lag times (duration) between data receipt and (query) feedback to the study sites are short. Only recent data is handled. CRF and query focus is about what’s currently happening with the subjects on the study sites. Earlier data collection is completed and new subjects and visits can be handled.
3. The study’s raw data listing is up to date. The amount of subjects and visits listed in the study’s raw data listing reflect the current number of subjects and visits conducted at the study sites.
4. Continuously, over 90% of data is clean. Reflecting an ongoing, up to date data verification and data cleaning process. This reveals that data is reviewed immediately after receipt for inconsistencies, and proper feedback (queries) to the study sites is communicated as soon as possible.
5. Communication about clinical data is mainly about study results; about meeting the study objectives, safety and efficacy objectives. In fact, focus shifts more and more to study content, because study conduct is under control.

The information that indicates if your clinical data collection and verification is doing fine is already available. You only need to find out where to get it for your study, and how to read and use it!

© 2011, Maritza Witteveen, ProCDM

You’re welcome to re-publish this newsletter if you add the following text to it. This is an article from Maritza Witteveen of ProCDM. Helping clinical research professionals who struggle with data handling, to get reliable trial data, so they can work on their other study goals. Receive tips and the free e-book ‘Five strategies to get reliable, quality clinical data’ by subscribing via http://www.procdm.nl/pages/knowledgebase.asp.

A clinical study is a structured way of collecting data on product safety and efficacy. … a structured way of performing research….
CRF design should follow this structure. However, sometimes you can get stuck during CRF design. Like I found myself struggling creating a CRF. 14 years of clinical data management experience and suddenly not being able to deliver a CRF for approval…
Hours and hours of work spent on CRF design. Each page modified at least five times. Updated for the data to be collected, the sequence of collection, as well as minor textual and lay-out updates.

Why didn’t CRF design progress?

1. A major baseline measurement differed in collected CRF fields (data) as compared to the comparative measurement at other (follow-up) visits. Start and end dates collected at baseline while recording modifications only for follow-up.

2. The CRF collected incomplete information for a measurement. Data collected during visits of which, at study discontinuation, no complete picture could be made, because end dates were not asked for.

These two causes had such an impact that, although everyone involved wanted, no one was able to declare the CRF ready for approval. Until, with external support, the incomplete information was revealed.

After discovery we then re-designed CRF pages to capture complete information. While doing that, it turned out that we even became able to re-use all repetitive measurements for visits. Which comforted and strengthened the spirit that the CRF could soon be signed for approval.

100% complete and re-usable data collection. Within four days after discovery of incomplete information, the CRF was finalized for the CRF fields (data captured) and lay-out. It took another week to adjust, test and document the data checks for correct queries popping up.

Did we noticed these incorrect basic CRF design requirements at the start, we could have saved a lot of time and energy. In fact, we did notice that something wasn’t right, but we couldn’t point it.

Solutions to progress CRF design:

1. This experience got me again on the track that a CRF should ALWAYS have look alike CRF pages for the same measurements captured at different time-points. No matter how complex the study design is. A clinical study is a structured way of doing research. And the CRF should reflect this structure, amongst others, through repetitive measurements.

2. Secondly, I’ve seen that if a measurement is collected, the complete measurement should be collected. Even if it seems that only part of the measurement is needed to answer the clinical study objectives. For each measurement, collect at least a clear start and end date, the result, the collection method and the clinical outcome. If any of these are missing, you’ll get stuck in providing a complete picture of what happened at the study site(s).

However, better stuck in CRF design progress, than capturing real study data with a bad designed CRF….

© 2011, Maritza Witteveen, ProCDM

You’re welcome to re-publish this newsletter if you add the following text to it. This is an article from Maritza Witteveen of ProCDM. Helping Clinical Research Directors, who struggle with clinical data management, to get reliable, quality data successfully. Receive tips and the free e-book ‘Five strategies to get reliable, quality clinical data’ by subscribing via http://www.procdm.nl/pages/knowledgebase.asp.