Data Reporting Problems Shelley Hewerdine & Khrystyna Valkiv Istanbul, 23rd March 2015 #EBMT2015 www.ebmt.org
Format of this session Recent EBMT Registry work Improving Data Quality Data reporting Quiz Your questions 2
How the data is used The Registry provides a pool of data to perform retrospective studies, assess epidemiological trends, or prepare prospective trials. Collected data on >500,000 transplants since they began in the 1970s. For basic and clinical Research Standardisation and Quality Control Accreditation (JACIE) Requirements - Health Authorities and Commissioners EBMT Membership requirements: MED-A for every transplant performed 3
Importance of good data Data as accurate and complete as possible for research and quality management Missing data and missing follow ups can skew data or lead to a statistical bias Follow up: very important part of the Registry, research helping to improve quality of life, long term. Example of centre reporting follow up only when something happens (Centre A) and a centre systematically reporting follow ups (Centre B): 4
Overall survival for Centres A and B 1.00 0.75 0.50 Centre B 0.25 Centre A 0.00 0 10 20 30 Years 5
1.00 Length of follow up Centres with same outcome may look worse when follow up is shorter 0.75 0.50 0.25 =< 5 years >10 years >5 and =<10 years 0.00 0 20 40 60 Years 6
Working to improve the data Tidy Up Your Data campaign. Have run a full list of data quality queries and sent to centres. (ProMISe users encouraged to run them to check data) 7
Tidy Up Your Data: list of report queries Diagnosis labelled as other or unknown (Main Classification only) Diagnosis sub classification labelled as other or unknown Best Response_Status after HSCT is Null and_or Unknown Disease status at HSCT is Null and_or Unknown Missing number of Disease status at HSCT No relapse prior to 2nd or subsequent HSCT Cause of death is GvHD but no GvHD recorded Cause of death is relapse but no relapse recorded Last assessment is HSCT assessment Patient alive with a cause of death Hospital or Unit unique patient number (UPN) missing Hospital unique patient number (UPN) shared by two patient records Patient date of birth missing Possible duplicate entry of same patient Allogeneic HSCT without a donor Source of stem cells missing 8
Working to improve the data: further requests Emailed all centres with a list of patients that have not had a follow up reported for 5 years or more. Aim to send these more regularly: twice a year Missing MED-A: requests sent to centres sending in data forms. Aim to widen requests to target more ProMISe users Duplicate registrations: many emails have been sent to query possible duplicates in centres and clean up the doubles Thank you for helping with our data quality requests 9
Quiz Quiz handouts 8 questions. Have a look in pairs or groups of 3 for approx 15-20 minutes 10
Quiz: Solutions 1. Which of the following should NOT be reported as a transplant registration? Answer b) Autologous reinfusion after graft failure This is not considered as a transplant. Check transplant definitions at the beginning of the MED-AB manual. Ask clinician or registry helpdesk if unsure. A boost is an infusion of cells from the same donor without conditioning, with the same donor being present in a proportion higher than 10% If cells are not from same donor OR there is conditioning OR donor cells are present at a lower proportion than 10%, then it is a genuine transplant. Please also report patients who die between conditioning and transplant. We enter the day 0 info and put the planned HSCT date as the date of death 11
Quiz: Solutions 2. A patient received Rituximab as a treatment for EBV or CMV reactivation, but you can t find this in the MED-A form, what should you do? This treatment is not part of the EBMT data collection. Can be entered in the database by advanced users, but unlikely to be analysed. Centres can record data for own use if preferred Recommendation: keep to MED-AB. And if you report MED-A but find you want to record MED-B items, report MED-B! 12
Quiz: Solutions 3. A patient has a rare diagnosis and you cannot find the right disease classification sheet or code. What do you do? Answer: b) Check the disease classifications list on the EBMT website If cannot find or unsure, answer c) Contact the registry helpdesk Answer a) Enter it in the database as uncoded is a no-no. It is very rare for diseases to be uncoded. Only enter an uncoded diagnosis if we advise you to. 13
Quiz: Solutions 4. In 2005 the percentage of transplant registrations with the best response at 100 days missing or unknown was 12%. Do you think this percentage has increased or decreased for recent registrations (2013 transplants)? Answer: decreased. The number of registrations with this item missing/unknown in 2013 was 7% Improvement over time: number of missing MED-A items decreased: 14
EBMT Registry Missing data items - average per HSCT Med-A submission 400 350 300 Number of centres 250 200 150 100 < 5 missing items 5-10 missing items >10 missing items 50 0 2011 2012 2013 2014 90% 80% 70% Percentage of centres 60% 50% 40% 30% 20% 10% 0% 2011 2012 2013 2014 < 5 missing items 5-10 missing items >10 missing items 15
Quiz: Solutions Remember you can run a report on the best response missing item, plus display records without disease status at transplant (and number of the status) properly filled. Crucial data quality reports on identifiers: Unique UPN per patient per centre Possible duplicates: all transplants need to go in one single patient record, no matter where performed or the number of transplants (Request form to access original transplant in other centre) Ensure date of birth and patient gender are filled 16
Quiz: Solutions 5. If a patient has more than one transplant, when should follow up be reported? Answer: b) Between transplants, then annually after the most recent transplant We will try to improve our data forms to help capture the follow up between transplants. Again a DQ report: No relapse prior to 2nd or subsequent HSCT 17
Quiz: Solutions 6. Donor Lymphocyte Infusions (DLI): Must all the infusions of the same regimen (given for the same indication within 10 weeks) be recorded separately, or must only the 1st infusion of the regimen be recorded with the total number of infusion in this regimen (without entering details for each infusion)? Answer: only the 1 st infusion of each 10 week cycle needs to be reported, giving the total number of infusions 18
Quiz: Solutions 7. What is the best way to send electronic data files to EBMT? E.g. scanned forms, HLA reports, Excel files with study or data queries Correct answer: b) send by email with password protection or c) use secure download facility in Promise. c) ideally best, available Promise 3. Answer a) Just send by email is a no-no for files containing identifiers, but continues to happen. 19
Quiz: Solutions 8. MED-A follow up form: discrepancies Cause of death is relapse, but relapse data not recorded. (Also may happen with gvhd). DQ reports available for both No date of last contact (also may happen with Excel requests) 20
Conclusions Positive response to Tidy Up Your Data and 5 Year follow up Decrease in number of missing items in MED-A Cleaning the duplicate registrations and shared UPNs. <400 possible doubles. Shared UPNs halved. Thank you for all your help and cooperation with the data, and the quality! 21
Your Questions? 22
Related Sessions EBMT 2015 Please complete feedback form.. then join us for a drink! Data Managers Reception Room 3B10 Tuesday 09:00 09:40 Changes to MED-A Please complete feedback forms 10:00 12:00 (computer room) HLA Data Entry 11.45 13:00 Introduction to HSCT and the MED-A form 14:00 15:30 (computer room) Avoiding & solving problems in ProMISe 14:45 15:25 Data Managers & Literature 23