Skip to main content

Table 2 Concepts and applied validation outcomes for n = 2 hospitals from the DTD

From: Validation of multisource electronic health record data: an application to blood transfusion data

Validation Approach Application
Order Concept Aim Outcome Average of two hospitals
External
 1 Concordance with report Data are concordant with (annual) report % agreement between number of products in annual blood bank report and DWH 98.7% (RBC 99.2%, PLT 97.6%, FFP 98.7%)
 10 Concordance with literature Data are concordant with previous findings in literature Comparison of distribution of blood products by age and gender per product type in the Netherlands Distributions were quite similar, only platelet use has shifted towards younger patients (Additional file 1: Figure S2.4).
 11 Concordance with experts Data are concordant with expert opinions; findings can be explained in a clinical context Plausibility of changes in Hb after blood transfusion The experts concluded that the plausibility is acceptable; the 1% unexpected decreases might be explained by other factors.
 12 Concordance with other databases Findings are concordant with other databases Comparison of findings with SCANDAT, a Scandinavian transfusion database The SCANDAT database has similar external concordance, completeness and linkage rates.
Internal
 2 Linkage data sources within DWH Entities occurring in multiple data tables can be linked % transfusions linked to issued products by id of the end product 99.96% (no link for n = 46 RBC, n = 5 PLT, and n = 1 FFP)
   % products issued linked to transfusion (indicates spilling rate) 97.65% (RBC 97.95%, PLT 99.25%, FFP 93.35%)
   % products that can be linked to donation(s); % products linked to donors Initially 96.73%, after improving the donation numbers this increased to 99.99%; the link from product to donor was 99.98%
 3 Identity No duplicates % duplicated transfusions (donation identification code + product type) 0.14% (initially this was 1%; it turned out that most duplicates were split products. Due to unavailable product codes in one hospital, the broader product type had to be used)
   % duplicated donations (donation identification code + product code) 0.005% (RBCs); 0% (FFP and PLT)
   % duplicated procedures codes 0% (all duplicates were removed, because it was expected that double registration would occur)
 4 Completeness No missing variables or values % patient ID; date of birth; gender, procedure date; Hb and thrombocyte counts; product code 100%; 99.99%; 99.99%; 100%; 99.8% and 97.5%; 50%
   % non-missing values for donor ID; date of birth; gender, Hb value, Expiration or Production date 100%; 99,995%; 100%; 98.8%; 100%
   % of transfusions that fall within at least one diagnosis start and end date 98% (see Additional file 1: Table S2.1)
 5 Uniformity Measures across time and data sources all have the same units, level of detail and/or coding system % product codes that occur in the reference list of ISBT product codes 50% (for one hospital product code was not available)
   % Diagnosis codes that occur in the reference list (of national diagnosis codes and descriptions) 96.1%
   % of Hb measurements from hospitals and blood bank with the same level of precision >99.6% uses 1 significant decimal in all sources
 6 Time patterns No unexplained changes over time Compare number of donations, products and donors of subsequent (calendar) years The observed decrease (Additional file 1: Figure S2.1) is in line with the known nationally decreasing trend.
   Examine number of transfusions per year per product type The relatively high decrease for FFP use (Additional file 1: Figure S2.2) can be explained by the introduction of ROTEM, a hemostasis testing method.
   Examine linkage percentage of transfusions to products issued per year In 2010 relatively many unlinked transfusions occurred (see Additional file 1: Figure S2.3). After blood bank data from the previous year 2009 was included, the linkage percentage increased to 99.8% or higher for all years.
 7 Plausibility Data are free of identifiable errors % donation date < date of pooling 100%
   % within limits for number of donations per donor per year (maximum is 3 (females) or 5 (males) for whole blood and 23 for plasma) FFP 100%; WB 99.8% (0.2% exceeds the limit with in total 6 or 7 donations within a year)
   % donor age > 18 and >70 years (minimum and maximum age for donating) 100% (only 0.0006% was >70 and 0.0004% was <18 and these were mainly autologous donations)
   % transfusion with increase (and decrease) in Hb level (Hb values + − 1 day around transfusion; difference > + − 8.8% is considered a clinical change) 54% increases; 6% decreases; 40% no change. Of those decreasing, 97% had a diagnosis indicating high bleeding risk
   % patient age < 121 years 100%
   Maximum number of transfusions per year Max tr. per year 476 (mainly FFP) for diagnosis TTP.
   % correct gender for Gynecology diagnoses 100%
   % patients with transfusions/ surgery after date of death 0.0% (n = 2 changed mortality status to NA)
   % with admission date before discharge date) (zero-length rule) 100%
   % with non-negative difference between expiration and transfusion date 99.93%
 8 Event attributes All attributes relevant to an event description are present % of pooled products that are linked to the correct number of unique donors (in this case 5 or 6 donors contribute to one pooled platelet product) 100%
   % of patients that are transferred to another hospital according to the ‘discharge destination’ variable 6%
   % transfusions linked to hospitalization (indicates outpatient transfusions) 99.16% (of which 23.64% day admissions, likely including transfusions given at the outpatient ward)
 9 Consistency hospitals within DWH No unexplained differences between hospitals Comparison of (validity) outcomes of the hospitals The two hospitals have very similar validity outcomes, not requiring further investigation.