Incoherences or atypical values in the french basis of hemovigilance "unknown aetiologies" - ISBTS 2007

Afssaps

Description

Blood and blood products
12/05/2007

Informations

• Publié par: ansm-blood-and-blood-products
• Publié le: 12 mai 2007
• Nombre de lectures: 15
• Licence : En savoir +
  Paternité, pas d'utilisation commerciale, pas de modification
• Langue: English

Extrait

XVII Regional Congress of ISBT, EUROPE Madrid 24th-27th June 2007 INCOHERENCES OR ATYPICAL VALUES IN THE FRENCH BASIS OF HEMOVIGILANCE unknown aetiologies" " M-Phuong VO Mai: Haemovigilance unit; Cyril Caldani: chief of the Haemovigilance unit; Beatrice Willaert: Haemovigilance unit; Imad Sandid: labile blood products evaluation; Jean-François Legras: responsable of the coordination of blood products; Pierrette Zorzi: chief of biological products department.

INTRODUCTION The adverse reactions of "unknown" aetiologies are rather benign reactions of the transfusion (94%) with an imputability which is often possible (78%). It is pro- bably for this reason that they remain until now insufficiently exploited, although they count for 9% of the set of the undesirable effects until 2004. Moreover since this date, the percentage has more than doubled, as far as concerning the accu- rateness, the consistency and the quality of the national data base of the French haemovigilance, which has more than 87500 notifications (1994-2006). Therefore, the question that arises is to understand the reasons of these anoma- lies and incoherencies, while studying the temporal series of adverse effects with unknown aetiologies. An immediate reaction is classified as "unknown" aetiology if: all assessments made come back negative, not any of the proposed diagnostic orientations from the formula can be kept, the effect is insufficiently informed in order to have a conclusion. OBJECTIVE The problem of detection of the incoherencies or atypical values is very old. It deals with all analysis on real data. It is necessary to detect these anomalies or rare events before beginning analysis of the data base, for theoretical and prac- tical reasons. Therefore, the "unknown" aetiologies of the national data base of the French haemovigilance have been analysed in order to answer to the questions: what is the quality of the data, what level of unknown reports can be accepted, how to take into account the restrictive factor of these observations in the analysis, is it necessary to eliminate these observations, to give them a discriminating weight or to use robust methods which hold account of them? METHODOLOGY, DATA AND TOOLS Method : •Main questions to solve before the application of a detection method: - What is the distribution of the initial data (frequency, graphic…) - When data are non-coherent: what distance/standard to choose? - When events are rare: what analysis to adopt? - What statistics or what test to adopt for those data? •Main questions to discuss after the application of a detection method: - If an observation is non-coherent or atypic, what is the reason? (Incoherent, extreme, or aberrant nature of the observation...) - How to treat these observations? (Correction, modification, elimination or integration in a particular model…) Data: The data come from the national basis "e-fit basis" of the patient's adverse events declared in France during the selected period 2000-2006 (e-fit: the French network reports system). Tools: Analysis are achieved with SAS. RESULTS •Non-coherent data: frequency The "unknown" aetiologies represent a frequency of 1 reaction for 3645 units of labile blood products in the period of [2000; 2003] and 1 reaction for 1580 units of labile blood products in the period of [2004; 2006]: most of incoherencies have been noted since 2004, beginning of the laun- ching up of the new electronic tool of reports e-fit.

Unknown aetiologies"

•What is the reason? The graph below shows that the number of the "unknown" notified by the correspondents of transfusion establishments tends to decrease according to the evolution of the trimesters, while those of the correspondents of heath establishment seem to take over.

Observations from ES - Observations from EST

From May 2004, the haemovigilance correspondents of the health establish- ments can report their undesirable effects to e-fit. However, the type of establishments where the reactions are notified and the type of hospital services where transfusions are achieved don’t seem to be the main explanation. •How to treat these observations? - Identification of atypical regions If these incoherencies are a matter for lack of notifications bound to the notifications with e-fit, they should meet in all regions. It is in fact the case as the graph below can show; nearly all regions are concerned, with more or less marked differences. Therefore, a careful attention can be put on the regions, for which the gaps compared to a standard (2.4) would be the most important then.

Progression of "unknown" aetiologies - ratio 2004-06 and 2000-03

- Analysis of particular clinical pictures or singulars Another correction of the data which is also possible could be based on the analysis of the "unknown" reactions having common clinical pictures. Presently, a clinical picture of thrill or/and fever in exclusion of all other clinical sign can be associated close to NHFR (non haemolytic febrile reac- tions). If the 2 times series "unknown" reactions and NHFR are jointly corrected, we obtain 2 smoothing trends.

CONCLUSION

Non-coherent data Initial data "Unknown aetiologies and NHFR"

Corrections New data "New class of unknown aetiologies and NHFR"

Besides, another issue could come out after examining the "unknown" reactions having divergent clinical pictures from a standard. This approach seems to be conclusive because, compared to the other diagnoses, the unknown diagnosis (excluding clinical picture of the shiver or fever previously seen) give to a myriad of clinical signs (until 7 signs by report). As the icterea, fevers and/or thrills, pains are however reminiscent of the "unknown" aetiologies; it is convenient to analyse the other reports and signs as atypical figures Unknown 2006, N=1450 Num b b y e  r r  e o p f o  s r i t gns Num A ber  % of R

2 3

443 237

6 10 7 1 1 450 Associations of most frequent signs Association Number of signs of AR

30.6% 16.3%

0.7% 0.1% 100%

% - - - - - - FR- - - - - - 126 8.7%          - AU- - - - - - - - - - - 62 4.3%  - - 58 4.0% - - - - - - - - - - - - AU- - - - FR- - - - - - 46 3.2%  - DO- - - - - - - - - 41 2.8% - - - - - - - - - - HT- - - - 23 1.6% - - - - - FIFR- - - NA- - 21 1.4% - - - DO- FIFR- - - - - - 21 1.4% - AU- - - - - - - - NA- - 19 1.3% - AU- - - - - - HT- - - - 17 1.2% - AU- - - FIFR- - - NA- - 15 1.0%   Legend:Anguis='AG'; Other= 'AU'; Schok='CH'; Pain ='DO'; Dyspnae=DY'; Fever='FI'; Chills='FR'; Hyperbilirubinery='HB'; Hypotension='HT'; Icteria='IC'; Nausea='NA'; Acute lung oedema ='OE'; Oligoanuria='OL'; Diffuse haemorage='SY' Research of atypical values by a multifactor function - Another method is: 1-to calculate a representative equation of the time serie of the "unknown" obser- vations, the adjusted better (method of the least squares by examples) then 2-to eliminate the aberrant values beyond the confidence interval of 95%. 1-Function ("unknown") above: f(x) when x = ∑ i Xi with i= 0, 1, …= signs and blood components: Variable y "unknown" endogens of binary type y = 0, 1 Variables xi: blood components=hexogen variable of ordinary type; signs=hexogen variable of binary type Analysis of estimations maximum of vraisemblance ParameterDF  Pr 2 of Wald > Khi 2Estimation Erreur std Khi INTERCEPT  1.47191 - 58.3318 0.1927 .0001 < BLOOD COMPONENTS 1 - 9.0545 0.0026 0.1198 0.0398 SHIVERING  .0001 < 0.0748 75.01271 0.6480 FEVER  .0001 < 29.1829 0.07421 0.4008 URTICARIA  < 126.5924 .00011 - 0.2432 2.7365 NAUSEA / GNIMITOV  43.6623 < .00011 0.8120 0.1229 ANGUISH 1 0.3723 0.1492 6.2305 0.0126 PAIN  125.6978 0.12251 1.3735 .0001 < SCHOCK  0.0057 0.3700 7.65541 1.0237 DYSPNEA  6.7695 0.00931 0.3371 0.1296 ACUTE LUNG ŒDEMA 1 -1.3723 17.0434 0.3324 .0001 < OTHER 1 0.6367 61.7443 0.0810 .0001 <

In an approximate way, the points above the value 4 represent the aberrant values or the evaluations of observations for which the residues are big. 3-Tests value Estimatio Odds ratios ns Parameter Limits 95%Point estimate confidance of Wald BLOOD COMPONENTS 0.887 0.820 0.959 SHIVERING 1.912 1.651 2.214 FEVER 1.493 1.291 1.727 URTICARIA 0.065 0.040 0.104 NAUSEA / NIGMOTIV 2.252 1.770 2.866 ANGUISH 1.451 1.083 1.944 PAIN 3.949 3.106 5.021 SCHOCK 2.784 1.348 5.749 DYSPNEA 1.401 1.087 1.806 ACUTE LUNG ŒDEMA 0.254 0.132 0.486 OTHER 1.690 1.613 2.215 Association of estimated probabilities and observed responses Percent Concordant76.5Somers' D0.575 Percent Discordant19.0Gamma0.602 Percent Tied4.4Tau-a0.178 Pairs7444768c0.788 The estimation of Odds ratios is interesting: percent concordant=76.5, percent discordant=19.0. The curve of Rock draws c=78.8%: i.e. the statistical c, which associates the "pair's observations/estimations", shows a sensitivity of 0.788 and vice versa a specificity of 0.212.

This analysis tends to demonstrate that: -The temporal distribution of the increase of "unknown" reactions after 2004 is closely due to the method of reporting and the profile of the people who declare the reports. -The first tests of modelling according to the syndromes of these aetiologies and blood components are not sufficiently robust and other tests have to be done for further checks. -The most important is to have reliable statistics of quality in order to achieve epidemiological studies in an adequate way.