PMWS development in pigs from affected farms in Spain and Denmark

1. PMWS development in pigs from affected farms in Spain and Denmark Anders Stockmarr Statistics Section Dept. of Applied Mathematics and Computer Science Technical University of Denmark Symposium in Applied Statistics January 29, 2013

2. Enteredprojectgroupafter all experiments had beenperformed

3. PostweaningMultisystemicWastingSyndrome(PMWS) • Multifactorialsyndromefor pigs. • Not a ‘disease’, but an immune system breakdown. • Clinicalsigns: • Weightloss; • Enlargedlymphnodes; • Respiratorydistress; • Sometimes diarrhea and jaundice; • Death/’wasting’. • VERY costly (Armstrong and Bishop 2004); fattening pigs that do not put on weight or die are of courseproblematic. • Cause: Unknown. Associated with PorcineCirvovirustype 2 (PCV2), but the exact association is not clear.

4. Cause of PMWS and thisstudy • In general, Unknown (at least for whatregards PMWS). • Associated with PorcineCirvovirustype 2 (PCV2), but verydifficult to reproduce in controlled studies with PCV2 infectionsalone.

5. Cause of PMWS and thisstudy • In general, Unknown(at least for whatregards PMWS). • Associated with PorcineCirvovirustype 2 (PCV2), but verydifficult to reproduce in controlled studies with PCV2 infectionsalone. • Meta analysis (Thomás 2008) suggests that PMWS maybereproducedthroughinfection with PCV2 and co-infections with otherpathogens. • Thus, in the studythatthisanalysis is based on, welooked at measures for infections with PCV2 and the followingpathogens: • Porcineparvovirus; • Svine influenza virus, strainsH1N1 or H3N2; • Lawsoniaintracellularis; • Porcine Reproductive and Respiratory Syndrome virus, European and American variant; • Aujesky’s disease virus; • Mycoplasma hyopneumonia; • Salmonella Spp.

6. Purpose of Study Analysis • To uncover the role of specificpathogens in the development of PMWS WorkingHypotheses: • The development of antibodiestowardspathogensthroughseroconversionafterinfectionincreases the risk of developing PMWS. • Immunityinherited from the motheranimal has a reducingeffect on the risk of developing PMWS.

7. PMWS Diagnosis • Presence of compatible clinical signs • Moderate to severe lymphocyte depletion • Granulomatous inflammation in lymphoid tissues • Detection of moderate to high amount of PCV2 withintheselesions (Segalés et al., 2005; Sorden, 2000).

8. PMWS Diagnosis • Presence of compatible clinical signs • Moderate to severe lymphocyte depletion • Granulomatous inflammation in lymphoid tissues • Detection of moderate to high amount of PCV2 withintheselesions (Segalés et al., 2005; Sorden, 2000). Not possible to diagnose without an autopsy.

9. Data material • Antibodymeasurementsweretaken at pre-specified time points; • Animals wereselected in Denmark and Spain afterclinicalsigns (cases) and youthanized; • Age-matchedcontrolswereselected (fewer) and youthanized;

10. Data material • Antibodymeasurementsweretaken at pre-specified time points; • Animals wereselected in Denmark and Spain afterclinicalsigns (cases) and youthanized; • Age-matchedcontrolswereselected (fewer) and youthanized; • However, the ‘cases’ were not diagnosed at selection, as thisrequires an autopsy. • Some of the ‘cases’ turned out not to be PMWS diagnosed…

11. Data material • Antibodymeasurementsweretaken at pre-specified time points; • Animals wereselected in Denmark and Spain afterclinicalsigns (cases) and youthanized; • Age-matchedcontrolswereselected (fewer) and youthanized; • However, the ‘cases’ were not diagnosed at selection, as thisrequires an autopsy. • Some of the ‘cases’ turned out not to be PMWS diagnosed… • And some of the controlscouldturn out to be cases, had theybeenallowed to live on…

12. Solution: SurvivalAnalysis Framework • PMWS status at autopsy; death/failure is PMWS development, if not observations arecensored at autopsy. Wasting non-PMWS animalsexcluded but used for control. • Covariates: HerdID, and: • Longitudinalmeasurements of antibodytitres/ OD% for the followingpathogens on 135 pigs (DK), 120 pigs (E) : • Porcineparvovirus (PPV); • Porcinecirvovirus type 2 (PCV2); • Svine fluH1N1 or H3N2 (SIV); • Lawsoniaintracellularis (LAW); • European Porcine Reproductive and Respiratory Syndrome virus (PRRSV.E); • American Porcine Reproductive and Respiratory Syndrome virus (PRRSV.U).

13. Relations to WorkingHypotheses • No direct measure of time for seroconversion; • No direct measure of maternalimmunity (motheranimalscannotbeused due to cross-fostering).

14. Relations to WorkingHypotheses • No direct measure of time for seroconversion; • No direct measure of maternalimmunity (motheranimalscannotbeused due to cross-fostering). Construction of such measures necessary

15. Maternalimmunity • Not possible to usevalues for motheranimals due to cross-fostering: Pigletsaretaken from onemotheranimal and laid at another, to maximizepigletsurvival. • Maternalimmunityestimated as the maximumregistratedantibodymeasurement in the firstthreeweeks of life.

16. Seroconversion Times • Pathogenantibodymeasurementsdeclines with time, until an infectionmakes it rise again. • The time point for seroconversion is the point in time whereantibodyconcentrationincreasesafter the initial decline, withoutdelay.

17. Seroconversion Times • Pathogenantibodymeasurementsdeclines with time, until an infectionmakes it rise again. • The time point for seroconversion is the point in time whereantibodyconcentrationincreasesafter the initial decline, withoutdelay. • To estimatethisestimate from only a few observations, the antibodyconcentrationprogress is estimatedthrough regression of 2ndorderpolynomials on the longitudinaldata. • The Seroconversion Time is estimated as the time point corresponding to the vertex of the parabola.

18. Seroconversion and No Seroconversion

19. Seroconversion and No Seroconversion

20. Survival Analysis • P(PMWS case in [t:t+Δt) | no case at t) ≈ λ(t) Δt • Cox’ Proportional Hazardsmodel for animali: λi(t) = λ0(t)exp( β1Xi1(t)+β2Xi2(t)+…+βkXik(t) ) • Covariatesarematernalimmunity (not time dependent), seroconvertiontimes. and interactionswithin and betweenthese (time dependent). • λ0 is non-parametric and not modelled. P(PPV seroconvertedanimalcase in [t:t+Δt) |not case at t)/ P(non-seroconvertedanimalcase i [t:t+Δt) | not case at t) =exp( βppv) if all othercharacteristics match; • Relative risksonlybecauseλ0 is not modeled.

21. ExemplifiedHazardRatio Development

22. Sensitivity Analysis • In order to contemplate the impact of the built-in impreciseness of the estimates of seroconversions, a sensitivity analysis was carried out after model reduction. Gaussian noise was added to the seroconversiontimes considering that 95% of the new seroconversion times should be within one week of the original estimates. • Noise addition and model reductionwasperformed 20 times; • In order to be rendered trulysignificant a significant factor must appear in at leasthalf of the analyses’ final models.

23. Estimation Method – Few Data, ManyCovariates • Include all seroconversions and interactions. Reduce. • Includeeachmaternalimmunity in a pre-specifiedsequence, and all interactions with factors in current model. Reduce. • Include all apparently non-significant factors againthrough forward selection, one at a time. Reduce. • Repeat steps 1-3 untilnochangesresults; ienoincrease in Cox’ partiallikelihood.

24. Results • DK: • SeroconversionagainstLAW; • SeroconversionagainstPRRSVe; • maternalimmunityagainstPCV2; • maternalimmunityagainstLAW. • Spain: • MaternalimmunityagainstLAW, PCV2, PPV, PRRSV and SIV.

25. Results • DK: • SeroconversionagainstLAW; • SeroconversionagainstPRRSVe; • maternalimmunityagainstPCV2; • maternalimmunityagainstLAW. • Spain: • MaternalimmunityagainstLAW, PCV2, PPV, PRRSV and SIV.

26. Results

27. SignificantImpact? • Create an indexI for animals with maternalimmunityaround average, by differentiating the log of the Cox PH after the covariates (in distributionalsense for seroconversions) for factors interacting, and takemeans of these. For factors not interacting, the indexI is the parameter estimate.

28. Indexes; Values and Significances

29. Indexes;Impact

30. WorkingHypotheses • The development of antibodiestowardspathogensthroughseroconversionafterinfectionincreases the risk of developing PMWS. In CONTRAST to results for LAW • Immunityinherited from the motheranimal has a reducingeffect on the risk of developing PMWS. CONFIRMEDfor • PCV2, LAWin Denmark, • PCV2, PPV and SIV in Spain. CONTRASTED for • PRRSV in Spain.

31. PossibleExplanations • SeroconversiontowardsLAW: • We DON’T observeinfections but merelyseroconversions; • Animals maybeinfected but unable to seroconvert due to progressing immune defiency; • Thatanimalsseroconvertmayindicate a functioning immune system, whichovershadows the weakeningeffect of infection with Lawsoniaintracellularis. • Lack of seroconversioneffectsfor Spanish data: • Stronglyheterogeneouspopulation. • Maternalimmunity in Spanish data: • Consistent with litterature for PCV2, PPV, SIV (seerefs in paper). • PRRSV resultsmaybeexplained by the heterogeneouspopulation; thusmaternalimmunitymayindicatehighpresence of PRRSV which is known from the litterature as a PMWS trigger.

32. Spanish Data and Seroconversions • Difficult to identifyeffects from such distributions of seroconversions. • But the lack of time-dependent covariatesmeansthat it is sensible to comparerisks and frequenciesthroughgrouping. Theyagree…

33. Conclusion • Protectiveeffect of seroconversionagainstlaw (DK). • Protectiveeffects of maternalimmunityagainst PCV2 (DK,E), PPV (E) and SIV (E). • Aggravatingeffect of maternalimmunityagainst PRRSV. • All effectsmaybecompatible with present knowledgeexceptlaw, whereseroconversionincreases the riskwhenmaternalimmunity is low; ie the diseasetriggers PMWS unlessmaternalimmunity is high. First report on this. • The level of detail in the analysis is new compared to existingknowledge. • Care shouldbetakenwhengeneralizingspanishresults due to population heterogeneity. • Furtherworkshouldinclude PCR data to counterindirectdetection of infections.