Latent models in the development and improvement of tools for health outcomes measurement

_______________________________________________________________________ PDSPPROGRAMA DOUTORAL EM SAUDE PÚBLICA UNIVERSIDADE DO PORTO FACULDADE DE MEDICINA Mil on Se e o Ba os da Sil a La en models in he de elopmen and imp o emen o ools o heal h ou comes measu emen Po o - 2012 _______________________________________________________________________ I Mil on Se e o Ba os da Sil a La en models in he de elopmen and imp o emen o ools o heal h ou comes measu emen Disse ação de candida u a ao g au de Dou o em Saúde Pública, ap esen ada à Faculdade de Medicina da Uni e sidade do Po o, ealizada sob a o ien ação cien í ica da P o esso a Dou o a Ana Ri a Pi es Gaio, do Depa amen o de Ma emá ica da Faculdade de Ciências da Uni e sidade do Po o e da co-o ien ado a P o esso a Dou o a Ana Aze edo Ca doso de Oli ei a, do Depa amen o de Epidemiologia, Medicina P edi i a e Saúde Pública da Faculdade de Medicina da Uni e sidade do Po o Po o 2012 _______________________________________________________________________ _______________________________________________________________________ II “All models a e w ong; some a e use ul.” Albe Eins ein Es e abalho oi ealizado com o apoio inancei o da Fundação pa a a Ciência e a Tecnologia e da Fundação As a Zeneca Po o 2012 _______________________________________________________________________ _______________________________________________________________________ III A .º 48º, § 3º - “A Faculdade não esponde pelas dou inas expendidas na disse ação.” (Regulamen o da Faculdade de Medicina da Uni e sidade do Po o – Dec e o-Lei nº 19337 de 29 de Janei o de 1931) _______________________________________________________________________ _______________________________________________________________________ IV Co po Ca ed á ico da Faculdade de Medicina do Po o P o esso es Ca ed á icos E e i os Dou o Manuel Albe o Coimb a Sob inho Simões Dou o Jo ge Manuel Me gulhão Cas o Ta a es Dou o a Ma ia Amélia Dua e Fe ei a Dou o José Agos inho Ma ques Lopes Dou o Pa ício Manuel Viei a A aújo Soa es Sil a Dou o Daniel Filipe Lima Mou a Dou o Albe o Manuel Ba os da Sil a Dou o José Manuel Lopes Teixei a Ama an e Dou o José Hen ique Dias Pin o de Ba os Dou o a Ma ia Fá ima Machado Hen iques Ca nei o Dou o a Isabel Ma ia Amo im Pe ei a Ramos Dou o a Deolinda Ma ia Valen e Al es Lima Teixei a Dou o a Ma ia Dulce Co dei o Madei a Dou o Al ami o Manuel Rod igues Cos a Pe ei a Dou o Rui Manuel Almeida Mo a Ca doso Dou o An ónio Ca los F ei as Ribei o Sa ai a Dou o José Ca los Ne es da Cunha A eias Dou o Manuel Jesus Falcão Pes ana Vasconcelos Dou o João F ancisco Mon eneg o And ade Lima Be na des Dou o a Ma ia Leono Ma ins Soa es Da id Dou o Rui Manuel Lopes Nunes Dou o José Edua do To es Ecken o h Guima ães Dou o F ancisco Fe nando Rocha Gonçal es Dou o José Manuel Pe ei a Dias de Cas o Lopes Dou o Manuel An ónio Caldei a Pais Clemen e Dou o An ónio Albino Coelho Ma ques Ab an es Teixei a Dou o Joaquim Adelino Co eia Fe ei a Lei e Mo ei a _______________________________________________________________________ _______________________________________________________________________ V P o esso es Jubilados ou Aposen ados Dou o Abel José Sampaio da Cos a Ta a es Dou o Abel Vi o ino T igo Cab al Dou o Alexand e Albe o Gue a Sousa Pin o Dou o Ál a o Je ónimo Leal Machado de Aguia Dou o Amândio Gomes Sampaio Ta a es Dou o An ónio Augus o Lopes Vaz Dou o An ónio Ca alho Almeida Coimb a Dou o An ónio Fe nandes da Fonseca Dou o An ónio Fe nandes Oli ei a Ba bosa Ribei o B aga Dou o An ónio Ge mano Pina Sil a Leal Dou o An ónio José Pacheco Palha Dou o An ónio Luís Tomé da Rocha Ribei o Dou o An ónio Manuel Sampaio de A aújo Teixei a Dou o Belmi o dos San os Pa ício Dou o Cândido Al es Hipóli o Reis Dou o Ca los Rod igo Magalhães Ramalhão Dou o Cassiano Pena de Ab eu e Lima Dou o Daniel San os Pin o Se ão Dou o Edua do Jo ge Cunha Rod igues Pe ei a Dou o Fe nando de Ca alho Ce quei a Mag o Fe ei a Dou o Fe nando Ta a ela Veloso Dou o F ancisco de Sousa Lé Dou o Hen ique José Fe ei a Gonçal es Lecou de Menezes Dou o José Augus o Fleming To inha Dou o José Ca alho de Oli ei a Dou o José Fe nando Ba os Cas o Co eia Dou o José Luís Medina Viei a Dou o José Manuel Cos a Mesqui a Guima ães Dou o Le i Eugénio Ribei o Gue a Dou o Luís Albe o Ma ins Gomes de Almeida Dou o Manuel Augus o Ca doso de Oli ei a Dou o Manuel Machado Rod igues Gomes Dou o Manuel Ma ia Paula Ba bosa Dou o a Ma ia da Conceição Fe nandes Ma ques Magalhães Dou o a Ma ia Isabel Amo im de Aze edo Dou o Má io José Ce quei a Gomes B aga Dou o Se a im Co eia Pin o Guima ães Dou o Valdema Miguel Bo elho dos San os Ca doso Dou o Wal e F ied ich Al ed Osswald _______________________________________________________________________ _______________________________________________________________________ VI Ao ab igo do A .º 8º do Dec e o-Lei nº388/70 azem pa e des a disse ação as seguin es publicações: I. Se e o M, Gaio R, Lucas R, Ba os H. Assessmen o he gene al public's knowledge abou heuma ic diseases: e idence om a Po uguese popula ion- based su ey. BMC Musculoskele Diso d. 2010;11(1):211. II. Se e o M, Gaio R, Lou enco P, Al elos M, Be encou P, Aze edo A. Indi ec calib a ion be ween clinical obse e s - applica ion o he New Yo k Hea Associa ion unc ional classi ica ion sys em. BMC Res No es. 2011;4:276. a. Se e o M, Gaio AR, Aze edo A. Calib a ion: e ec on he misclassi ica ion o NYHA. (submi ed). III. Se e o M, Gaio AR, Lou enço P, Al elos M, Gonçal es A, Lune N, Be encou P, Aze edo A. Diagnos ic alue o pa e ns o symp oms and signs o hea ailu e: applica ion o la en class analysis wi h concomi an a iables. (submi ed). a. Se e o M, Pe ei a M, Be encou P, Gaio R, Aze edo A. B-Type Na iu e ic Pep ide Measu ed in Se um – Calib a ion Using Plasma Samples o Resea ch Pu poses. Clinical Labo a o y. 2011; 57(11- 12):1015-9. b. Se e o M, Lopes C, Lucas R, Ba os H. De elopmen o a ool o he assessmen o calcium and i amin D in akes in clinical se ings. Os eopo os In . 2009;20(2):231-7. c. Se e o M, Gaio AR. A simple ool o ma ch a concomi an a iable la en class model classi ica ion. (submi ed). _______________________________________________________________________ _______________________________________________________________________ VII Jú i da P o a de Dou o amen o Dou o José Agos inho Ma ques Lopes (P esiden e) Faculdade de Medicina, Uni e sidade do Po o Dou o José Hen ique Dias Pin o de Ba os Faculdade de Medicina, Uni e sidade do Po o Dou o José Manuel Gonçal es Dias Ins i u o Uni e si á io de Lisboa Dou o a Ca la Ma ia de Mou a Lopes Faculdade de Medicina, Uni e sidade do Po o Dou o Ana Ri a Pi es Gaio Faculdade de Ciências, Uni e sidade do Po o Dou o a Helena C is ina de Ma os Canhão Faculdade de Medicina, Uni e sidade de Lisboa Dou o Paulo Jo ge da Sil a Noguei a Faculdade de Medicina, Uni e sidade de Lisboa _______________________________________________________________________ _______________________________________________________________________ VIII À Fe nanda e à Cla a Aos meus Pais _______________________________________________________________________ _______________________________________________________________________ 1 Resumo Resumo _______________________________________________________________________ _______________________________________________________________________ 2 P e endemos com es a ese comp eende o papel dos modelos de a iá eis la en es pa a desen ol e e melho a ins umen os de medição em saúde. Nes e con ex o, o am desenhados 3 es udos p incipais: I. Se e o M, Gaio R, Lucas R, Ba os H. Assessmen o he gene al public's knowledge abou heuma ic diseases: e idence om a Po uguese popula ion- based su ey. BMC Musculoskele Diso d. 2010;11(1):211. Es e es udo e e como obje i o iden i ica as c enças e os conhecimen os sob e doenças eumá icas numa amos a da população ge al e iden i ica g upos al o pa a educação pa a a saúde, a a és da aplicação de modelos de a iá eis la en es Pa icipan es o am selecionados du an e o seguimen o de uma coo e ep esen a i a da população adul a do Po o, Po ugal; 1626 pa icipan es comple a am um ques ioná io que incluíam i ens sob e conhecimen os ge ais de doenças eumá icas. Modelos de a iá eis la en es disc e as e con ínuas o am usadas pa a iden i ica alhas nos conhecimen os e g upos al o. O modelo de a iá eis la en es con ínuas iden i icou 2 dimensões: um elacionado com c enças (la en e 1) e ou o elacionado com ca ac e ís icas, a amen o e impac o das doenças eumá icas (la en e 2). O modelo de a iá eis la en es com 3 classes e inou es es esul ados: A p imei a classe ap esen a a baixa p obabilidade de ace a aos i ens associados com a p imei a la en e (média de 39%), a segunda classe ap esen ou baixa p obabilidade de ace a aos i ens da segunda dimensão (média de 62%). A e cei a classe ap esen ou uma p obabilidade al a de ace a odos os i ens (média de 79%). II. Se e o M, Gaio R, Lou enco P, Al elos M, Be encou P, Aze edo A. Indi ec calib a ion be ween clinical obse e s - applica ion o he New Yo k Hea Associa ion unc ional classi ica ion sys em. BMC Res No es. 2011;4:276. Es e es udo e e po obje i o calib a o sis ema de classi icação NYHA en e di e en es obse ado es, aspi ando aumen a a sua iabilidade. Resumo _______________________________________________________________________ _______________________________________________________________________ 3 Os 265 indi íduos, de um o al de 1136 adul os esiden es no Po o, Po ugal, com idade ≥ 45 anos, que epo a am al a de a esponde am a um ques ioná io com 4 i ens pa a ca ac e iza a g a idade dos sin omas. O ques ioná io oi aplicado po 7 médicos, que ambém classi ica am a capacidade uncional do indi íduo de aco do com a NYHA. Cada sujei o oi a aliado po um médico. A classi icação NYHA pelo oi calib ada com o mé odo conco en e, usando um modelo de aço la en e de 1- pa âme o. Disc epâncias en e os obse ado es o am a aliadas po di e enças nos pon os de co e en e as classes NYHA I-II e II-III no ní el da a iá el de aço la en e. A a iá el de aço la en e es imada pelo modelo oi u ilizada pa a p e e a classi icação NYHA pa a cada obse ado . O ní el da a iá el de aço la en e pa a o p imei o e segundo pon o de co e a iou pa a cada obse ado de -1,92 a 0,46 e 1,42-2,30, espe i amen e. A conco dância en e a a iá el de aço la en e es imada e classi icação dos obse ado es NYHA oi de 88% (kappa = 0,61). a. Se e o M, Gaio AR, Aze edo A. Calib a ion: e ec on he misclassi ica ion o NYHA. (Subme ido) Es e es udo em como obje i o mos a que a calib ação do sis ema de classi icação da NYHA en e di e en es obse ado es aumen a a sua alidade, eduzindo a má classi icação. Os 265 indi íduos, de um o al de 1136 adul os esiden es no Po o, Po ugal, com idade ≥ 45 anos, que epo a am al a de a esponde am a um ques ioná io com 4 i ens pa a ca ac e iza a g a idade dos sin omas. O ques ioná io oi aplicado po 7 médicos, que ambém classi ica am a capacidade uncional do indi íduo de aco do com a NYHA. Cada sujei o oi a aliado po um médico. A classi icação NYHA oi calib ada com o mé odo conco en e, usando um modelo de aço la en e de 1- pa âme o. Compa amos a á ea sob a cu a ROC (AUC) e o alo p edi i o da e são calib ada com a e são não calib ada do NYHA pa a p e e a p esença de uma sé ie de medidas obje i as do ecoca diog ama da unção ca díaca. A AUC mos ou um aumen o da capacidade p edi i a do NYHA após a calib ação, em g ande medida pelo aumen o da azão de e osimilhança do NYHA I. Resumo _______________________________________________________________________ _______________________________________________________________________ 4 III. Se e o M, Gaio AR, Lou enço P, Al elos M, Gonçal es A, Lune N, Be encou P, Aze edo A. Diagnos ic alue o pa e ns o symp oms and signs o hea ailu e: applica ion o la en class analysis wi h concomi an a iables. (Subme ido). O p opósi o dos au o es oi iden i ica pad ões de sin omas e sinais baseado em dados ecolhidos na p á ica clinica de o ina, e a alia o seu alo de diagnós ico, endo em con a a p obabilidade a p io i de IC. Baseados num es udo ans e sal o am a aliados mil e cen o e quinze pa icipan es da comunidade com idade≥45 do Po o, Po ugal, em 2006-2008. Fo am iden i icados pad ões u ilizando a análise de classes la en es, usando a iá eis concomi an es pa a p e e a classe a que cada pa icipan e pe ence. Os pad ões usa am 11 sin omas e sinais, ab angendo sob eca ga de olume e hipope usão. Sexo, idade, educação, obesidade, diabe es e his ó ia de en a e do miocá dio ou IC o am incluídos como a iá eis concomi an es. A solução com 3 pad ões oi supo ada pelo c i é io de in o mação Bayesiano: 10,1% dos pa icipan es ap esen a am um pad ão com sin omas de al a de a e sob eca ga de olume (pad ão 1), 27,8% ap esen a am um pad ão ca ac e izado p incipalmen e po sob eca ga de olume (pad ão 2) e 62,1% e am essencialmen e assin omá icos (pad ão 3); o melho ajus e do modelo e i icou-se quando incluímos as a iá eis concomi an es. A azão de e osimilhanças pa a os pad ões 1, 2 e 3 pa a a dis unção sis ólica do en ículo esque do oi 3,4, 1,1 e 0,6, e pa a a dis unção dias ólica do en ículo esque do oi 3,5, 1,4 e 0,5, espe i amen e. a. Se e o M, Pe ei a M, Be encou P, Gaio R, Aze edo A. B-Type Na iu e ic Pep ide Measu ed in Se um – Calib a ion Using Plasma Samples o Resea ch Pu poses. Clinical Labo a o y. 2011;11 Es e es udo e e como obje i o a alia a p ecisão das medições de BNP em so o usando um ensaio imuno luo omé ico pa a p e e as concen ações de BNP em plasma e classi ica os indi íduos de aco do com os pon os de co e habi uais. Fo am incluídos 27 indi íduos com idade mínima de 45 anos, pa icipan es de um es udo de coo e Po uguês. Amos as de sangue o am ecolhidas em ubos de Resumo _______________________________________________________________________ _______________________________________________________________________ 5 plás ico de sangue o al, con endo um ácido e ilenodiamino e acé ico pa a ob e plasma ou a i ado de coágulo pa a ob e so o. O loga i mo na u al de so o BNP oi calib ado com o loga i mo na u al de plasma BNP usando uma equação de eg essão linea . Os pa âme os de eg essão es imados o am 0,58 (IC 95%: 0,23-0,93) pa a β0 e 1,01 (IC 95%: 0,90-1,11) pa a β1. A conco dância absolu a en e o plasma BNP e que p e is os pela equação de aco do com os pon os de co e 30 e 100 pg/mL o am 96,3% (kappa = 0,92) e 96,3% (kappa = 0,91), espe i amen e. b. Se e o M, Lopes C, Lucas R, Ba os H. De elopmen o a ool o he assessmen o calcium and i amin D in akes in clinical se ings. Os eopo os In . 2009;20(2):231-7. Es e es udo e e como obje i o o desen ol imen o de uma e amen a pa a medi o consumo alimen a de cálcio e de i amina D em Po ugal, e a e i a u ilidade de a iá eis não alimen a es. En e is ado es einados ecolhe am in o mação de 2414 adul os da cidade do Po o, Po ugal, a a és de um ques ioná io de equência alimen a (QFA) semies u u ado. Fo am selecionados pa a e amen a os alimen os com maio con ibuição pa a o consumo e as a iá eis não alimen a es (sexo, idade, e índice de massa co po al (IMC)). Di e en es ap oximações es a ís icas o am usadas pa a p e e o consumo. O g á ico de Bland-Al man oi usado pa a compa a as p e isões da e amen a com os esul ados do QFA. Os i ens selecionados pa a p e e o consumo de cálcio o am o lei e (38%), o queijo (12%), o iogu e (10%) e o sexo; pa a a i amina D, o peixe go do (39%), enla ado (9%) e mag o (7%), os o os (5%), a ca ne e melha (5%), a idade e o IMC. O g á ico de Bland–Al man mos ou que a média das di e enças oi de 0,0 (limi es de conco dância= [-220.67; 220.77]) mg/dia e 0,0 (limi es de conco dância = [-1.03; 1.05])µg/dia, espe i amen e pa a o cálcio e i amina D. c. Se e o M, Gaio AR. A simple ool o ma ch a concomi an a iable la en class model classi ica ion (subme ido) Resumo _______________________________________________________________________ _______________________________________________________________________ 6 O p opósi o des e es udo é eplica a classi icação ob ida pela análise de classes la en es (LCA) com a iá eis concomi an es usando e amen as simples de es a ís ica. Baseados num es udo ans e sal o am a aliados mil e cen o e quinze pa icipan es da comunidade com idade≥45 do Po o, Po ugal, em2006-2008. Fo am iden i icados pad ões u ilizando a análise de classes la en es, usando a iá eis concomi an es pa a p e e a classe a que cada pa icipan e pe ence. Os pad ões usa am 11 sin omas e sinais, ab angendo sob eca ga de olume e hipope usão. Sexo, idade, educação, obesidade, diabe es e his ó ia de en a e do miocá dio ou insu iciência ca díaca (IC) o am incluídos como a iá eis concomi an es. Análise de classes la en es iden i icou 3 classes com di e en es pe is clínicos, que chama-mos “pad ão de IC sin omá ico”, “pad ão de sob eca ga de olume” e “pad ão assin omá ico”. De inimos pon uação o al como o núme o de sin omas e sinais obse ados, e a pon uação concomi an e como o núme o espe ado de sin omas e sinais p e is os a a és de uma eg essão binomial nega i a ealizada sob e pon uação o al e usando as a iá eis concomi an es. Cada pon uação concomi an e oi es imada a a és do sis ema de pon os ou uma égua ci cula baseada na in o mação das a iá eis concomi an es. A á o e de classi icação com a pon uação o al e concomi an e oi usada pa a p e e classi icação dos indi íduos p e is os pela análise de classes la en es. A conco dância absolu a en e as classes p e is as e o iginais oi de 89,7% (in e alo de con iança de boo s ap 95% (B95%CI) =(88,0; 91,6)) e o espe i o Kappa oi de 0,802 (B95% CI = (0,781; 0,850)). Resumo _______________________________________________________________________ _______________________________________________________________________ 7 As p incipais conclusões da in es igação são as seguin es: 1. O uso de modelos la en es aplicados a es as escalas especí icas pe mi iu a iden i icação de di e en es dimensões e de g upos-al o ele an es na população ge al. 2. Os pon os de co e da classi icação da NYHA en e obse ado es o am bas an es disc epan es e a calib ação conco en e a a és dos modelos de aço la en e pode se usada pa a calib a o g ande núme o de obse ado es na mesma escala, con ibuindo pa a minimiza o p oblema de iabilidade da classi icação da NYHA. Es e ipo de ap oximação pode se ú il pa a minimiza a a iabilidade em ou as classi icações baseadas nas pe ceções de doen es e/ou médicos. a. A me odologia de calib ação pode se ú il pa a melho a a alidade da classi icação do NYHA na p á ica clínica e em con ex o de in es igação, con ibuindo pa a minimiza a má classi icação nes a escala. 3. O uso de a iá eis concomi an es pode melho a o alo de diagnós ico dos pad ões de sin omas e sinais e, consequen emen e, aumen a a u ilidade dos sin omas e sinais an o pa a o diagnós ico como em medidas de espos a. a. As amos as de so o não podem se usadas pa a es ima os alo es absolu os de concen ação de plasma, mas os alo es de BNP medidos em so o e calib ados podem se usados pa a classi ica de o ma co e a os indi íduos endo em con a os pon os de co e habi uais. b. As equações es imadas pelo melho modelo de p e isão de consumo alimen a de cálcio e de i amina D pe mi i am desen ol e um so wa e e uma égua ci cula ú il no con ex o clínico. c. Uma e amen a simples pa a o diagnós ico de insu iciência ca díaca oi desen ol ida que pe mi i á a u ilização no con ex o clínico dos cuidados p imá ios onde em ge al não exis em disponí eis de imedia o medidas obje i as de unção e de es u u a ca díaca. _______________________________________________________________________ _______________________________________________________________________ 8 Abs ac Abs ac _______________________________________________________________________ _______________________________________________________________________ 9 This hesis aimed o unde s and he ole o la en models in he imp o emen and de elopmen o heal h ou comes measu emen . In his con ex , h ee main esea ch ques ions we e add essed: I. Se e o M, Gaio R, Lucas R, Ba os H. Assessmen o he gene al public's knowledge abou heuma ic diseases: e idence om a Po uguese popula ion- based su ey. BMC Musculoskele Diso d. 2010;11(1):211. This s udy aimed o iden i y he inco ec belie s and common knowledge abou heuma ic diseases in a sample o he gene al popula ion and o iden i y a ge g oups o heal h educa ion – applica ion o la en models. Pa icipan s we e selec ed du ing he ollow-up o a ep esen a i e coho o adul popula ion o Po o, Po ugal; 1626 pa icipan s comple ed a ques ionnai e ha included gene al knowledge i ems abou heuma ic diseases. Disc e e and con inuous la en a iable models we e used o iden i y knowledge laws and he a ge g oups. A con inuous la en a iable model iden i ied wo dimensions: one ela ed o gene al belie s (la en 1) and ano he conce ning cha ac e is ics, ea men and impac o heuma ic diseases (la en 2). A 3-class la en a iable model e ined hese esul s: he i s class p esen ed he lowes p obabili ies o co ec answe o i ems associa ed wi h he i s la en (mean o 39%), and he second class p esen ed he lowes p obabili ies o co ec answe o i ems wi h he second la en (mean o 62%). The hi d class showed he highes p obabili y o a co ec answe o almos all he i ems (mean o 79%). II. Se e o M, Gaio R, Lou enco P, Al elos M, Be encou P, Aze edo A. Indi ec calib a ion be ween clinical obse e s - applica ion o he New Yo k Hea Associa ion unc ional classi ica ion sys em. BMC Res No es. 2011;4:276. This s udy aimed o calib a e he NYHA classi ica ion sys em be ween di e en obse e s, aspi ing o inc ease i s eliabili y. Abs ac _______________________________________________________________________ _______________________________________________________________________ 10 Among 1136 communi y-dwelle s in Po o, Po ugal, aged ≥45 yea s, 265 epo ing b ea hlessness answe ed a 4-i em ques ionnai e o cha ac e ize symp om se e i y. The ques ionnai e was adminis e ed by 7 physicians who also classi ied he subjec ’s unc ional capaci y acco ding o NYHA. Each subjec was assessed by one physician. We calib a ed NYHA classi ica ions by he concu en me hod, using 1-pa ame e la en ai model. Disc epancies be ween obse e s we e assessed by di e ences in abili y h esholds be ween NYHA classes I-II and II-III. The abili y (s anda d no mal a iable) es ima ed by he model was used o p edic he NYHA classi ica ion o each obse e . Es ima es o he i s and second h esholds o each obse e anged om -1.92 o 0.46 and om 1.42 o 2.30 s anda d de ia ions o abili y, espec i ely. The ag eemen be ween es ima ed abili y and he obse e s’ NYHA classi ica ion was 88% (kappa=0.61). a. Se e o M, Gaio AR Aze edo A. Calib a ion: e ec on he misclassi ica ion o NYHA. (Submi ed) P e ious s udies showed an in e -obse e ag eemen o he NYHA classi ica ion o app oxima ely 55%. Thus, he calib a ion o he NYHA classi ica ion sys em be ween di e en obse e s is expec ed o inc ease i s alidi y, educing misclassi ica ion. A he s anda dized clinical in e iew subjec s who epo ed o ha e b ea hlessness (n=265) we e p esen ed o a 4-i em ques ionnai e on unc ional capaci y o cha ac e ize he se e i y o symp oms. The ques ionnai e was adminis e ed by 7 physicians who also classi ied he subjec ’s unc ional capaci y acco ding o NYHA. Each subjec was assessed by one physician. Calib a ion o NYHA classi ica ion ac oss each se o indi iduals assessed by each physician was pe o med by he concu en me hod using he ou pa ien i ems as ancho i ems. We es ima ed he a ea unde he ROC cu e (AUC) and likelihood a io using he calib a ed and non-calib a ed NYHA class I e sus II-III o p edic he p esence o a se ies o objec i e s uc u al o unc ional ca diac abno mali ies as assessed by echoca diog aphy a es . The a ea unde he ROC cu e (AUC) o NYHA class o p edic he ou comes conside ed showed an o e all imp o emen in disc imina ion o NYHA class a e i s calib a ion, la gely a he expense o he likelihood a io o NYHA I. In oduc ion _______________________________________________________________________ _______________________________________________________________________ 17 independen . Whene e J equals 1, i.e., he la en ai is 1-dimensional, he analysis is e e ed o as i em esponse heo y. The mo e gene al si ua ion co esponding o J g ea e han 1 is e e ed o as mul idimensional. An assump ion o LTM is local independence o he i ems, ha is, once he alues o he ai s a e known, he i ems mus be independen . Mo eo e , i he i ing o he model is good, hen i em pa ame e s can be es ima ed independen ly o he pa icula sample o indi iduals ha answe ed he ques ionnai e (wi hin a linea ans o ma ion) (10). This p ope y is known as i em pa ame e in a iance. In addi ion, an indi idual’s es ima ed ai is no dependen upon he pa icula sample o i ems chosen om he ba e y (10). This o he p ope y is known as abili y pa ame e in a iance. The abo e ea u es show al eady ha LTM o e mo e comp ehensi e and accu a e e alua ion o he psychome ic p ope ies o a gi en ques ionnai e o e classical measu emen echniques [7]. They also allow o an op imal sho ening o he ques ionnai e, when necessa y, and o he e alua ion o he pe o mance o he educed measu emen . Fu he , LTM p o ide an es ima e o he eliabili y o a scale along he whole ai (in o ma ion unc ion) (11), ins ead o a single es ima e o eliabili y as we ha e wi h C onbach’s alpha, o ins ance, in he classical measu emen heo y. The in o ma ion unc ion p o ides a g aphical ep esen a ion o he p ecision o he measu emen a each ai alue o ei he an indi idual es i em (i em in o ma ion unc ion, IIF) o he en i e es ( es in o ma ion unc ion, TIF). The g ea e he in o ma ion p esen a a gi en ai alue, he mo e p ecise and/o eliable he measu emen will be a ha alue (12). The in o ma ion unc ion is especially appealing wi h heal h ou comes, whe e i is impo an o maximize he eliabili y o a scale a he cu -sco e, and is e y use ul in cons uc ing sho o ms o ailo ed assessmen s, ensu ing ha he selec ed subse o i ems p o ides an adequa e eliabili y. Ano he impo an ea u e o LTM is ha hey ha e a ‘‘buil -in’’ linking mechanism (8). Linking is a gene al e m ha can be used o e e o bo h equa ing and calib a ion. In oduc ion _______________________________________________________________________ _______________________________________________________________________ 18 LTM de ine a scale o he unde lying la en a iable and he i ems a e calib a ed wi h espec o ha scale. The linking p ope y o LTM means ha , once he i ems a e calib a ed o a popula ion (i.e., i em pa ame e s a e known), compa able sco es on a gi en cons uc may be calcula ed o esponden s om ha popula ion who answe ed only a subse o he i ems, wi hou in e media e equa ing s eps. In applica ions whe e i em pa ame e s a e a p io i no known, he linking o wo o mo e scales is s ill ai ly s aigh o wa d p o ided bo h o ms measu e he same cons uc and he e a e some o e lapping i ems on he o ms. This means ha LTM can be used o equa e and calib a e a la ge numbe o i ems o di e en ques ionnai es; by doing so, we a e able o be e unde s and he s uc u e and o de o he domain-speci ic i ems o each o he , as well as he in e ela ions among i ems ac oss he abili y con inuum and o design compu e based ques ionnai es (13) . La en class models, also known as analysis o ini e mix u e models, assume ha he la en a iable is ca ego ical. This amewo k b ings se e al s a is ical ad an ages o e s anda d classi ica ion app oaches. Fi s ly, i allows p oblems such as he choice o he numbe o classes and o he classi ica ion me hod o be ecas as s a is ical model choice p oblems. Secondly, LCM can be po en ially imp o ed h ough he use o concomi an a iables, i.e., a iables ha in luence he p e alence o classes, hus pe mi ing he iden i ica ion o mo e p ecise ca ego ies. Finally, o gi en alues o he esponse and concomi an a iables, pos e io class membe ship p obabili ies o each indi idual a e p oduced (14). Examples o heal h amewo ks whe e la en class analysis was used include d ug abuse/dependence (15), alcohol use (16), ma e nal dep ession (17) and die a y pa e ns (18). The e is no ob ious choice be ween LTM and LCM. In he con ex o disease diagnosis he iew ha domina es is he ca ego ical one, because i mee s clinical needs and allows epo ing o heal h-ca e planne s, while in LTM i is di icul o ind na u al cu poin s o h esholds o he ai s, educing i s use ulness o p o ide a classi ica ion. In oduc ion _______________________________________________________________________ _______________________________________________________________________ 19 Howe e , LCM igno e possible wi hin-class he e ogenei y such as indi idual di e ences in se e i y. In he ollowing sec ions we a e going o ma hema ically desc ibe he la en models used in his hesis as well as i s applica ions o heal h ou comes measu emen . La en Models La en ai models La en ai models a e a class o la en models o which he la en a iables a e me ical. In applica ions o heal h, hey a e equen ly used in ques ionnai es, scales, es s,… ha ha e dicho omous o o dinal i ems. The mos commonly used 1-dimensional LTM o hese ypes o i ems a e p esen ed in able 1. Wi hin his esea ch p ojec , only he 2-pa ame e logis ic model and he g aded esponse model will be used in applica ions. The e o e, we b ie ly desc ibe hese wo me hods h oughou . Table 1. Desc ip ion o se e al commonly used 1-dimensional LTM Mod el I em esponse ypes Model cha ac e is ics 1-pa ame e logis ic model Dicho omous Disc imina ion powe cons ained o be equal ac oss i ems. Th esholds a y ac oss i ems 2-pa ame e logis ic model Dicho omous Disc imina ion and h esholds a y ac oss i ems 3-pa ame e logis ic model Dicho omous Includes pseudo-guessing pa ame e , besides uncons ained disc imina ion and h eshold pa ame e s G aded model O dinal Disc imina ion and h esholds a y ac oss i ems Nominal model Poly omous Disc imina ion and h esholds a y ac oss i ems Pa ial c edi model Poly omous Disc imina ion powe cons ained o be equal ac oss i ems. Th esholds a y ac oss i ems. Ra ing scale model Poly omous Disc imina ion powe and i em h eshold s eps cons ained o be equal ac oss i ems In oduc ion _______________________________________________________________________ _______________________________________________________________________ 20 Two-pa ame e logis ic model Two-pa ame e logis ic models o dicho omous esponses, y, assume ha he 1- dimensional la en a iable, z, is me ical and (s anda d) no mally dis ibu ed, he dis ibu ion o each i em condi ional on he la en a iable is binomial B(1, π j(z)), and he ela ionships be ween z and y can be desc ibed by an ogi e-shaped unc ion called an i em cha ac e is ic cu e (ICC), )( 01 1 1 )( jj z j e z ββ π −− + = (1) whe e π j(z) = P(yj=1|z), i.e., he p obabili y o co ec ly answe ing i em j gi en he alue o z, and βj0 and βj1 a e he pa ame e s o he model o be es ima ed. This model is simila o a logis ic eg ession; he e, howe e , independen a iables a e no obse able. Fo i em j, he pa ame e βj0 ep esen s he abili y alue a which he p obabili y o co ec ly answe ing he i em is 0.5, and is called di icul y pa ame e . This pa ame e is exp essed on he same scale as he ai . The pa ame e β1j is called disc imina ion pa ame e and ep esen s he slope o ICC nea he espec i e di icul y pa ame e , hus indica ing how well an i em disc imina es indi iduals wi h ai alue nea he di icul y pa ame e ; he highe he alue o he disc imina ion pa ame e , he highe is i s disc imina ion. The ICC can also be desc ibed by he ollowing pa ame e iza ion, which has he ad an age o immedia ely allow o he ex ension o he model o mo e han one la en a iable, )( 1 ' 0 1 1 )( z j j j e z ββ π +− + = (2) In oduc ion _______________________________________________________________________ _______________________________________________________________________ 21 The in e p e a ion o he disc imina ion pa ame e emains he same bu he (new) di icul y pa ame e , ' 0j β , has a di e en in e p e a ion om abo e; i now ep esen s he logi alue o an indi idual a mean alue o he ai . The absolu e alue o he disc imina ion pa ame e desc ibes he ela ion s eng h be ween he co esponden i em and he ai . An exac equi alence has been demons a ed be ween he disc imina ion pa ame e s o he LTM model and he s anda dized ac o loadings, i.e., co ela ion coe icien be ween he la en a iable z and he i ems unde lying he con inuous a iable (3). ( ) 2/1 2 1 1 1 1 *+ = j j j β β β G aded esponse model The g aded esponse models (GRM) a e an ex ension o bina y esponse models app op ia e o use when i em esponses a e o o dinal ype (19). These models assume ha he pe o mance o an indi idual on he i ems is explained by only one (s anda d no mal) a iable, deno e by z as abo e. In he g aded esponse models, each i em is desc ibed by a se o cu es, called i em ope a ion cha ac e is ic cu es (IOCC). The i em ope a ion cha ac e is ic cu e o ca ego y k o i em j ep esen s he p obabili y o endo sing ca ego ies highe han k condi ional on he alue o he subjec ’s ai . I is gi en by he equa ion )( )(01 1 1 )( kjj z j e kyP ββ −− + => (3) whe e y j deno es he esponse o i em j. I em ope a ion cha ac e is ic cu es o an i em wi h K ca ego ies a e cha ac e ized by K pa ame e s: he slope (disc imina ion), j1 β , which is he same o all ca ego ies o he In oduc ion _______________________________________________________________________ _______________________________________________________________________ 22 i em, and he h esholds (di icul y), )(0 kj β , which a e as many as he numbe o ca ego ies minus one. The h eshold pa ame e be ween wo ca ego ies ep esen s he abili y alue a which he p obabili y o indica ing he highes o hese wo o highe is 50%. So, he h eshold pa ame e s a e exp essed in he same scale as he abili y. The slope pa ame e indica es how well an i em is able o disc imina e indi iduals wi h abili y alues nea he espec i e h eshold. The slope pa ame e may also be in e p e ed as desc ibing how an i em may be ela ed o he abili y. The s eepe he slope he highe is he i em disc imina ion. La en class models Fo a ixed numbe , K, o classes, he model assumes ha he popula ion densi y is exp essed as a weigh ed ini e sum o K componen densi ies, 1,…, k, wi h pa ame e s θ1,.., θk, espec i ely, and each densi y is iden i ied wi h a class. Fo he indi idual i, le yi deno e he esponse ec o o obse a ions on J a iables. In ou applica ion, hese inpu a iables, also deno ed by i ems o mani es a iables, will be ei he dicho omous o o dinal. The gene al model is ( ) ( ) kik K kkii i θηψ α |,| 1),(| yxy x ∑ = = whe e ηk is he p obabili y o class k membe ship, and xi is a ec o o concomi an a iables ha in luences he p e alence o he classes h ough he pa ame e s α. The ec o ψ =(α, θ1,…, θK) is he se o model pa ame e s ha a e o be es ima ed. The model assumes ha ∑ = = K kk 1 1 η . Assuming independence o he coo dina e esponse ec o s yij wi hin each class k, and ha he mul i a ia e densi y k is he same ac oss classes, say =( 1,…, J), he model w i es In oduc ion _______________________________________________________________________ _______________________________________________________________________ 23 ( ) ( ) ∏ ∑ = = = J jjkijj K kkii i1 1),(| |,| θηψ α yxy x . The (sub)model ha akes accoun o he inpu a iables dis ibu ion wi hin each class is deno ed by componen speci ic model, and he (sub)model ha s udies he in luence o he concomi an a iables on he classes p e alence is called concomi an a iable model. The la e model will gene ally be a mul inomial logi model wi h pa ame e s α. Indi iduals a e assigned a class acco ding o he s anda d modal alloca ion om pos e io class membe ships. In his esea ch p ojec , LCM will be conside ed in wo si ua ions: o bina y i ems only, and o i ems o mixed-mode ype. The densi ies k a e conside ed acco dingly o he i ems’ ype. Challenges o he applica ion o la en models in heal h ou comes measu emen s The epidemiological ansi ion om in ec ious o ch onic diseases, e en in low- and middle income coun ies, is al eady well es ablished and is o majo ele ance o heal h planning. Musculoskele al diseases, ischemic hea disease and ce eb o ascula disease we e amongs he leading causes o mo bidi y in he wo ld in 2001 (20). Musculoskele al diseases cons i u e indeed a majo public heal h challenge o ou aging socie ies (21). P o iding he gene al popula ion and pa ien s wi h good quali y in o ma ion is an impo an s a egy o he managemen o ch onic diseases. Knowledge leads o changes in a i udes and beha iou s, and di ec ly in luences heal h s a us (22), and adequa e in o ma ion can p omo e sel -managemen skills necessa y o coping wi h he disease inc easing adhe ence o he apy (23). A ecen e iew (24, 25) iden i ied signi ican limi a ions and cons ain s in measu ing os eopo osis In oduc ion _______________________________________________________________________ _______________________________________________________________________ 24 knowledge as a single domain, as i should include mul i-dimensional aspec s like causes o isk ac o s, p e en ion, consequences and ea men . Hea ailu e is a complex clinical synd ome esul ing om a a ie y o s uc u al o unc ional ca diac diso de s. The diagnosis o hea ailu e (HF) equi es a compa ible clinical synd ome and demons a ion o ca diac dys unc ion by imaging o unc ional es s (26, 27). A clinical examina ion is always he i s s ep in a diagnos ic app oach o possible HF and u he in es iga ion is condi ional on ini ial clinical judgmen . Howe e , indi idual symp oms (such as dyspnoea and a igue) and signs (e.g. hi d hea sound and e idence o conges ion) a e gene ally un eliable and ha e limi ed alue o diagnosing hea ailu e (28, 29). Se e al mul idimensional c i e ia based on symp oms and signs ha e been de eloped o e decades in an a emp o s anda dize he clinical assessmen o hea ailu e (30-37). When pa ien s ini ially labeled as ha ing hea ailu e a e in es iga ed using objec i e assessmen c i e ia, only a ound one hi d a e conside ed o uly ha e hea ailu e (38, 39). Obesi y, un ecognized myoca dial ischaemia o pulmona y disease commonly lead o alse posi i e hea ailu e diagnoses (38). Addi ionally, i may be di icul o dis inguish pa hologic condi ions om me e physical decondi ioning associa ed wi h ageing. Mo eo e , he a ying subjec i e impo ance a ibu ed o symp oms jus i ies a sys ema ic associa ion be ween epo ed symp oms and emale gende and psychosocial cha ac e is ics, bo h among he heal hy and hose wi h ca diac dys unc ion. Gende , age, educa ion and obesi y a e majo de e minan s o symp oms and signs sugges i e o hea ailu e (40), beyond hei ole as isk ac o s o hea ailu e, and may accoun o alse posi i e and nega i e classi ica ion. Fu he mo e, he clinical judgmen is modi ied based on he a p io i likelihood o HF (41), depending mainly on his o y o HF o myoca dial in a c ion, and on s ong isk ac o s o such condi ions. In oduc ion _______________________________________________________________________ _______________________________________________________________________ 25 The New Yo k Hea Associa ion (NYHA) unc ional classi ica ion is one o he s eps needed o ob ain a good clinical examina ion o HF. The NYHA classi ica ion was designed o clinical assessmen o pa ien s by physicians in 4 classes (I, II, III o IV) on he basis o he pa ien ’s limi a ions in physical ac i i ies caused by ca diac symp oms (42). The NYHA classi ica ion is de i ed la gely by in e ence om his o y and/o obse a ion o he pa ien in ce ain physical ac i i ies, and occasionally by di ec o indi ec measu emen o ca diac unc ion in esponse o s anda dized exe cises. The class a clinician decides o assign a pa ien o depends on he clinician´s in e p e a ion o wha is “o dina y” physical ac i i y, “sligh ” and “ma ked” limi a ions. This esul s in a high in e -obse e a iabili y. P e ious s udies showed an in e -obse e ag eemen o he NYHA classi ica ion o app oxima ely 55% (43, 44). Consequen ly he use o NYHA classi ica ion as an ou come measu e in clinical esea ch is a he poo . T adi ionally, LTM and LCM a e used sepa a ely in he de elopmen o ques ionnai es and scales. As e e ed abo e, he e a e di e en ad an ages in using LTM and LCM o de elop heal h ou comes measu emen s, hus he combina ion o bo h models and consequen ly he combina ion o hei bes ea u es could e ine exis ing ins umen s; o example, using bo h models we could ob ain simul aneous wi hin class he e ogenei y and na u al cu poin s o p o ide a classi ica ion. La en models can be use ul in he s anda diza ion o clinical assessmen s. E en wi h aining, i may be un ealis ic o expec ha clinicians will necessa ily ob ain equi alen clinical examina ions, o he ex en ha in e -obse e a iabili y is a negligible issue. LTM ha e been ex ensi ely used o ques ionnai e and scale s anda diza ion bu no o clinical examina ions; hei applica ion in his ield could con ibu e o imp o e i s s anda diza ion. In oduc ion _______________________________________________________________________ _______________________________________________________________________ 26 Finally, as e e ed abo e, LCM can be po en ially imp o ed h ough he use o concomi an a iables. Concomi an a iable LCM could accoun o known de e minan s o he ele an clinical indings and he a p io i p obabili y o he condi ion. These models could help mimicking he easoning in clinical diagnosis, which depa s om an a p io i p obabili y ha in luences he inal clinical conclusion. Chap e 1 _______________________________________________________________________ _______________________________________________________________________ 33 Chap e 1 _______________________________________________________________________ _______________________________________________________________________ 34 Chap e 1 _______________________________________________________________________ _______________________________________________________________________ 35 Chap e 1 _______________________________________________________________________ _______________________________________________________________________ 36 Chap e 1 _______________________________________________________________________ _______________________________________________________________________ 37 _______________________________________________________________________ _______________________________________________________________________ 38 Chap e 2 Indi ec calib a ion be ween clinical obse e s - applica ion o he New Yo k Hea Associa ion unc ional classi ica ion sys em Chap e 2 _______________________________________________________________________ _______________________________________________________________________ 39 Chap e 2 _______________________________________________________________________ _______________________________________________________________________ 40 Chap e 2 _______________________________________________________________________ _______________________________________________________________________ 41 Chap e 2 _______________________________________________________________________ _______________________________________________________________________ 42 Subchap e 2.1 _______________________________________________________________________ _______________________________________________________________________ 49 Subchap e 2.1 _______________________________________________________________________ _______________________________________________________________________ 50 Subchap e 2.1 _______________________________________________________________________ _______________________________________________________________________ 51 Subchap e 2.1 _______________________________________________________________________ _______________________________________________________________________ 52 Subchap e 2.1 _______________________________________________________________________ _______________________________________________________________________ 53 Subchap e 2.1 _______________________________________________________________________ _______________________________________________________________________ 54 Subchap e 2.1 _______________________________________________________________________ _______________________________________________________________________ 55 _______________________________________________________________________ _______________________________________________________________________ 56 Chap e 3 Diagnos ic alue o pa e ns o symp oms and signs o hea ailu e: applica ion o la en class analysis wi h concomi an a iables. Chap e 3 _______________________________________________________________________ _______________________________________________________________________ 57 Chap e 3 _______________________________________________________________________ _______________________________________________________________________ 58 Chap e 3 _______________________________________________________________________ _______________________________________________________________________ 65 Chap e 3 _______________________________________________________________________ _______________________________________________________________________ 66 Chap e 3 _______________________________________________________________________ _______________________________________________________________________ 67 Chap e 3 _______________________________________________________________________ _______________________________________________________________________ 68 Chap e 3 _______________________________________________________________________ _______________________________________________________________________ 69 Chap e 3 _______________________________________________________________________ _______________________________________________________________________ 70 Chap e 3 _______________________________________________________________________ _______________________________________________________________________ 71 Chap e 3 _______________________________________________________________________ _______________________________________________________________________ 72 Chap e 3 _______________________________________________________________________ _______________________________________________________________________ 73 Chap e 3 _______________________________________________________________________ _______________________________________________________________________ 74 Subchap e 3.1 ___________________________________________________________________________ _______________________________________________________________________ 81 Subchap e 3.1 ___________________________________________________________________________ _______________________________________________________________________ 82 Subchap e 3.1 ___________________________________________________________________________ _______________________________________________________________________ 83 _______________________________________________________________________ _______________________________________________________________________ 84 Subchap e 3.2 De elopmen o a ool o he assessmen o calcium and i amin D in akes in clinical se ings. Subchap e 3.2 _______________________________________________________________________ _______________________________________________________________________ 85 Subchap e 3.2 _______________________________________________________________________ _______________________________________________________________________ 86 Subchap e 3.2 _______________________________________________________________________ _______________________________________________________________________ 87 Subchap e 3.2 _______________________________________________________________________ _______________________________________________________________________ 88 Subchap e 3.2 _______________________________________________________________________ _______________________________________________________________________ 89 Subchap e 3.2 _______________________________________________________________________ _______________________________________________________________________ 90 Subchap e 3.3 ___________________________________________________________________________ _______________________________________________________________________ 97 Subchap e 3.3 ___________________________________________________________________________ _______________________________________________________________________ 98 Subchap e 3.3 ___________________________________________________________________________ _______________________________________________________________________ 99 Subchap e 3.3 ___________________________________________________________________________ _______________________________________________________________________ 100 Subchap e 3.3 ___________________________________________________________________________ _______________________________________________________________________ 101 Subchap e 3.3 ___________________________________________________________________________ _______________________________________________________________________ 102 Subchap e 3.3 ___________________________________________________________________________ _______________________________________________________________________ 103 Subchap e 3.3 ___________________________________________________________________________ _______________________________________________________________________ 104 Subchap e 3.3 ___________________________________________________________________________ _______________________________________________________________________ 105 Subchap e 3.3 ___________________________________________________________________________ _______________________________________________________________________ 106 ___________________________________________________________________________ _______________________________________________________________________ 113 Discussion Discussion ___________________________________________________________________________ _______________________________________________________________________ 114 This hesis aimed o unde s and he ole o la en models in he imp o emen and de elopmen o heal h ou comes measu emen . In his con ex , h ee main esea ch ques ions we e add essed: 1. The e inemen o a knowledge ques ionnai e abou heuma ic diseases di ec ed owa ds he gene al popula ion, he iden i ica ion o belie s and knowledge abou hese diseases, and he de ec ion o a ge g oups o heal h educa ion. 2. The calib a ion o he NYHA classi ica ion sys em be ween di e en obse e s, aspi ing o inc ease i s eliabili y and alidi y. 3. The ole o concomi an a iable la en class models in he diagnosis o HF, and how o ansla e he esul ing classi ica ion o he clinical p ac ice. La en models, bo h ai and class, p o ided he e idence base o he iden i ica ion o knowledge domains in he gene al popula ion ega ding heuma ic diseases. Addi ionally, his me hod was ins umen al o iden i y ele an a ge g oups o educa ional p og ammes. The i s iden i ied la en ai was associa ed wi h he ollowing s a emen s, which p obably ep esen w ong gene al belie s: “ heuma ic diseases a e mo e equen in olde women”, “ heuma oid a h i is is caused by poo die ”, “cold and damp wea he ”, “ heuma ic pa ien s should es and mo e as li le as possible” and “ heuma ic diseases canno be cu ed”, and “all heuma ic pa ien s end up in wheelchai s”. I ems abou ae iology, ea men and impac o heuma ic diseases we e ela ed wi h he second inden i ied la en ai , which was he e o e deno ed speci ic knowledge. The la en 3-class model showed ha 25.7% o he indi iduals ag eed wi h he alse gene al belie s bu did ha e speci ic knowledge abou he diseases, 30.8% did no ag ee wi h he gene al belie s and did no ha e speci ic knowledge, and 43.5% did no ag ee wi h gene al belie s and had speci ic knowledge. Discussion ___________________________________________________________________________ _______________________________________________________________________ 115 O e all his mean ha almos 60% o he indi iduals had some gaps in he o e all knowledge abou heuma ic diseases. The e we e di icul ies ega ding he iden i ica ion o whe he diseases whe e heuma ic (ankylosing spondyli is and ib omyalgia) o no (glandula e e and mul iple scle osis), and mo e han i y pe cen belie ed ha people wi h heuma ic diseases canno be cu ed and canno die om hose illnesses. The la e inding is simila o ha epo ed by o he s: a s udy wi h women aged 65-90 yea s in Canada (24) showed ha only 36% ag eed ha heal h p oblems caused by os eopo osis can be li e- h ea ening and ano he s udy ca ied ou in US adul s (45) ound ha only 63% co ec ly espond “ alse” o he s a emen “no medica ions can ea os eopo osis”. This s udy con i med ha i is impo an ha educa ional p og ams abou heuma ic diseases should be cen ed in he eldes and low educa ed indi iduals, in o de o coun e ac w ong gene al belie s oge he wi h he ob ious imp o emen o he iden i ica ion o he di e en heuma ic diseases. Speci ic educa ion p og ams o heal h p o essionals could also imp o e he communica ion o he speci ic knowledge o pa ien s and hei ela i es. The applica ion o LTM o he NYHA classi ica ion has objec i ely indica ed he main eason why se e al s udies ha e epo ed low in e -obse e eliabili y and consequen limi ed use ulness o ha classi ica ion as an ou come measu e. The main poin was he exis ence o disc epan h esholds be ween obse e s in he de ini ion o NYHA class I, II and III. Al hough he obse e s in he s udy we e expe ienced physicians wi h aining in he managemen o hea ailu e, he e we e s ill disc epancies be ween hei (subjec i e) e alua ions. In a-obse e eliabili y is e y impo an o in e p e changes in NYHA class in he indi idual pa ien who is assessed epea edly by he same physician. Ne e heless in e -obse e a iabili y should be o special conce n when pa ien s a e assessed by di e en physicians. This is pa icula ly impo an , in p ac ice, in unscheduled isi s o he clinic o he eme gency depa men , whe e pa ien s a e no assessed by hei usual a endan . These unscheduled isi s a e usually due o wo sening symp oms and an inc ease in he NYHA class, in Discussion ___________________________________________________________________________ _______________________________________________________________________ 116 compa ison wi h he p e ious clinical s a e, is used as a c i e ion o clinical decisions such as hospi al admission and in ensi y o he apy adjus men such as he use o in a enous medica ion. The e o e, in each se ing he NYHA classi ica ion was o be used, i would be use ul o iden i y he di e ences be ween he obse e s’ assessmen s and o calib a e hei classi ica ions. The absolu e ag eemen be ween he p edic ions o he NYHA classi ica ion ac oss obse e s was 65%. The same s a is ic was imp o ed o 88% a e compa ison o he ai p edic ions wi h he obse e classi ica ion, o each obse e . This showed how he subjec i i y o he h esholds a ec ed he eliabili y o he NYHA classi ica ion. A he same ime his imp o emen con i med he quali y o he ob ained calib a ion. Calib a ion me hodology can be use ul o imp o e he eliabili y be ween obse e s in clinical p ac ice and esea ch se ings. In clinical p ac ice, i is possible o use he ela ion be ween ancho i ems and abili y o s anda dize he classi ica ion among obse e s, and o gi e guidelines o imp o e he in e -obse e s eliabili y. In he esea ch se ing, he scale de ined by he ancho i ems and he ope a o s classi ica ion can be used as a s anda dized NYHA classi ica ion ha minimizes he subjec i i y o each obse e classi ica ion. In o de o e alua e he alidi y o he NYHA calib a ion, we assessed i s p edic i e alue wi h espec o a se ies o objec i e s uc u al o unc ional ca diac abno mali ies as assessed by echoca diog aphy a es . The disc imina ion powe o he calib a ed NYHA was shown o be be e han i s co esponden non-calib a ed scale, la gely a he expense o he likelihood a io o NYHA I. The majo limi a ion o his calib a ion s udy is i s small size. Whe eas i was sligh ly la ge han he minimum numbe equi ed o p ope ly i a 1-PL la en ai model, i was no la ge enough o allow o he applica ion o a 2-PL model (11); indeed we ob ained uns able i em pa ame e s and la ge s anda d e o s. Each indi idual was assessed by one obse e only, opposed o he ideal si ua ion whe e ha indi idual should be assessed by all obse e s. Discussion ___________________________________________________________________________ _______________________________________________________________________ 117 We do no hink o his as a limi a ion. Fi s , he e we e no s a is ically signi ican di e ences among he g oups e alua ed by each obse e ega ding sex, clinical his o y, sys olic blood p essu e, educa ion and le en icula sys olic dys unc ion; only age, body mass index and dias olic blood p essu e showed signi ican small di e ences. On he o he hand, he ancho i ems we e ela ed o each obse e ’s NYHA classi ica ion; so e en i he e we e an obse e g oup ha was disc epan om he o he s, he ancho i ems would gua an ee a good calib a ion. The e o e we a e con iden ha his appa en limi a ion did no ha e a majo impac on he esul s. The s udy add essing he las esea ch opic succeeded in iden i ying h ee pa e ns o synd omic agg ega ion o symp oms and signs o HF. Based on indings ou inely collec ed in cu en clinical p ac ice, we applied concomi an a iable LCM o accoun o known de e minan s o he ele an clinical indings and he a p io i p obabili y o he condi ion. These models we e shown o be use ul o s anda dize and quan i y he p obabilis ic easoning in clinical diagnosis, upon which decisions o u he in es iga ion and e en ea men need o be made. Mos o he ob ained likelihood a ios we e small, showing ha he pa e ns gene a ed ela i ely small changes om p e- o pos - es p obabili y o ca diac abno mali ies. On one hand, his was expec ed, conside ing ha many o hese abno mali ies a e known o be asymp oma ic in a la ge p opo ion o pa ien s o a long ime, and he symp oms a e unspeci ic. On he o he hand, i is compa ible wi h p e ious quan i ica ions o he diagnos ic alue o symp oms and signs (46). The concen a ion le el o B- ype na iu e ic pep ide is an es ablished bioma ke o he diagnosis o HF. The ob ained pa e ns showed a good diagnos ic pe o mance o exclusion o high BNP alues. Howe e hese concen a ions we e no measu ed acco ding o he manu ac u e ’s ecommenda ions. The manu ac u e ’s ecommends he use o whole blood o plasma specimens o measu e BNP and we ha e measu ed i in se um. Ne e heless, in a Discussion ___________________________________________________________________________ _______________________________________________________________________ 118 pilo s udy we es ablished a calib a ion equa ion o allow o he use o se um BNP and o he classi ica ion o he communi y indi iduals wi h he usual cu -o s. Fo he i s ime, o ou knowledge, he ob ained HF classi ica ion in eg a ed ac o s ha ha e a la ge impac on he p e alence o symp oms and signs sugges i e o HF. The no el y in ou applica ion, wi h espec o p e ious HF classi ica ions, is ha class p obabili ies a e adjus ed o concomi an a iables. Speci ically, he model es ima es he inc ease o dec ease in class p obabili ies o indi iduals condi ional on he espec i e concomi an a iables alues, con ibu ing o an inc ease in he disc imina ion and o a dec ease in he numbe o alse nega i es and alse posi i es. The inclusion o hese a iables imp o ed he i ing o he model. The disc imina ion powe o he concomi an a iable LCM was shown o be be e han ha o i s co esponden LCM, la gely a he expense o he likelihood a io o pa e n 3 (no symp oms and signs) o exclude ca diac abno mali ies. These esul s objec i ely indica e ha he use o concomi an a iables can imp o e he diagnos ic alue o he symp oms and signs pa e ns and, consequen ly, imp o e he use ulness o he symp oms and signs o diagnosis and as an ou come measu e. The po en ial o applica ion in o he se ings o complex diagnoses is e y high. The pa e ns iden i ied by his me hodological app oach depend on he ype o popula ion being s udied. In his s udy, in a sample o he gene al popula ion, h ee pa e ns we e iden i ied: “symp oma ic HF”, “symp oms and signs o conges ion” and “no symp oms and signs”. In a p e ious s udy, using a simila app oach in subjec s discha ged a e myoca dial in a c ion o acu e hea ailu e, he au ho s we e able o dis inguish di e en pa e ns (non- cases, hea ailu e and ad anced hea ailu e) (37). The pa e ns iden i ied in he cu en s udy a e app op ia e o diagnosis in he gene al popula ion only. The low p e alence o ad anced and se e e HF cases is a limi a ion o his s udy and could ha e unde es ima ed he disc imina i e capaci y o his se o i ems. Also, he p e alence o mo e speci ic symp oms and signs such as a hi d hea sound (47) was oo low in his sample o be able o ake hem in o accoun . The p oposed pa e ns a e likely o be mo e Discussion ___________________________________________________________________________ _______________________________________________________________________ 119 sensi i e bu less speci ic han p e iously a ailable sco es such as he F amingham c i e ia, suppo ing hei use ulness as po en ial sc eening ools o ini ial clinical in es iga ion ha do no aim o eplace ull in es iga ion in he clinical se ing. Clinical p ac i ione s who a e no esea ch scien is s may ha e di icul ies in he implemen a ion o complex s a is ical models in hei p ac ice ou ine, such as concomi an a iable la en class analysis. The p esen s udy aims a ansla ing he alida ed pa e ns in o a classi ica ion sco e, using an app oach o making complex s a is ical models use ul o p ac i ione s and esea che s, such as a ci cula ule (48) o poin s sys em (49), which was used o example o de elop he widely used F amingham isk sco es. The use o his ool will allow he iden i ica ion o high- isk candida es o hea ailu e who a e likely o ha e a subs an ial yield o posi i e indings when es ed o objec i e measu es o ca diac dys unc ion in clinical p ac ice. In addi ion, i will con iden ly exclude hea ailu e in o he s, hus o ien ing he clinical in es iga ion owa ds al e na i e di ec ions. Such a ool could also inc ease he disc imina ion and dec ease he numbe o alse nega i es and alse posi i es in epidemiological s udies on HF. The s udy succeeded in de i ing a ool ha will enable gene al p ac i ione s and in e nis s o diagnose hea ailu e in p ima y ca e, whe e in gene al objec i e measu es o ca diac s uc u e and unc ion a e no a ailable. The ollowed app oach was simila o ha o a p e ious one, whe e we designed a ool o es ima e he calcium and i amin D in akes, om a simpli ica ion o a complex s a is ical model. The ci cula ule main ad an age is ha i can be used when no compu e is a ailable o among people wi hou special compu e skills. In he HF con ex , he ool in eg a ed a ci cula ule and a able on he back. The e was a small ade-o be ween model simplici y (ci cula ule and able) and p edic i e accu acy (la en class analysis wi h concomi an a iables). The quali y o he ag eemen was excellen . Ne e heless, his ag eemen can be o e es ima ed once he ool was applied o only one sample o indi iduals. Discussion ___________________________________________________________________________ _______________________________________________________________________ 120 Mo eo e , he es ima ed eg ession coe icien s and ee h esholds can a y ac oss di e en subpopula ions. We ha e used boo s apping o ci cum en his ques ion; esul s showed a small bias o he es ima es. ___________________________________________________________________________ _______________________________________________________________________ 121 Conclusions Conclusions ___________________________________________________________________________ _______________________________________________________________________ 122 The applica ion o la en ai and la en class me hodologies p o ided he e idence base o he iden i ica ion o knowledge domains in he gene al popula ion ega ding heuma ic diseases. The me hod was ins umen al o iden i y ele an a ge g oups o educa ional p og ammes. I was shown ha he e we e se e al knowledge laws abou heuma ic diseases. One ou o ou indi iduals conside ed alse gene al belie s as ue, and app oxima ely 30% did no ha e de ailed knowledge on heuma ic diseases. Highe educa ion and he p esence o disease con ibu ed posi i ely o he o e all knowledge. La en models can also be use ul in he s anda diza ion o clinical assessmen s, namely in he iden i ica ion o di e en h esholds o di e en obse e s. In he NYHA classi ica ion, hose cu -o s end o be disc epan as he e is some subjec i i y inhe en o he classi ica ion. Concu en calib a ion h ough la en ai models can be used o link a la ge numbe o obse e s o he same scale. I p o ides a way o maximize he eliabili y and o minimize he misclassi ica ion o a classi ica ion. Classical la en class analysis can be adap ed o include e ec s on he p e alence o he di e en pa e ns, in a me hodology called concomi an a iable la en class analysis. These models can be use ul o s anda dize and quan i y he p obabilis ic easoning in clinical diagnosis, upon which decisions o u he in es iga ion and e en ea men need o be made. In he hea ailu e amewo k, concomi an a iables no di ec ly ela ed o clinical indings we e used o accoun o a p io i p obabili ies o he condi ion. We succeeded in iden i ying h ee pa e ns o synd omic agg ega ion o symp oms and signs o hea ailu e, based on indings ou inely collec ed in cu en clinical p ac ice. Rela i ely o he classical app oach, alidi y was imp o ed. A simple diagnos ic ool o gene al p ac i ione s and in e nis s was de eloped. I will enable hem o diagnosis hea ailu e in p ima y ca e, whe e in gene al objec i e measu es o ca diac s uc u e and unc ion a e no a ailable.