The fall risk (FR) and a related injury increase with age and with the association of neurological diseases (Parkinson’s Disease (PD) or stroke). Falls represent a fearsome event for an elderly for traumatic and psychological consequences, and costs are becoming unsustainable. Hence, fall prevention in elderly at risk is a public health priority. All international guidelines recommend removing the modifiable FR factors and implementing effective interventions on people at risk. Nevertheless, investigators have not used consistent classifications for FR factors, so using the WHO Family of International Classifications can be the more natural and logical solution to cover the lack of a universal reference framework. FR screening is the first component of effective fall prevention programs. To date, despite the use of numerous FR assessment tools, it is not possible to detect and predict elderly fallers with optimal diagnostic accuracy. The aims of the thesis were: 1) to validate a FR serial screening algorithm with high diagnostic accuracy in a sample of community-dwelling elderly, also with PD and stroke, in the prediction of at least one, two, and three falls in the following twelve months; 2a) to assess the neurological disease's effect on FR tests; 2b) to validate an ICD& ICF core set for FR in the same population. Using data collected in the PRE.C.I.S.A. trial on fall risk, we performed the following analyses: 1) to validate a FR screening algorithm we calibrated the VAE, VOE1 and VOE2 scales with classical and Rasch analysis and we calculated the two FRAT-up; we studied the diagnostic accuracy of single tools and screening algorithms, obtained with serial combinations of the scales and the two FRAT-up tools, and with logistic regression models, in the prediction of the described outcomes; we compared their properties for external validation; 2a) to assess neurological disease’s effect on the tools we conducted a Differential Item Functioning (DIF) analysis for the scales and a t-test for the two FRAT-up; 2b) to validate the core set we reviewed the FR factors and we linked them to classification categories. The available sample from the PRE.C.I.S.A. trial was constituted by 768 older adults (female 65.3%; mean age 75.8). From 29 observed variables, we calibrated three measurement scales (VAE, VOE1, and VOE), which showed a satisfactory fit to the Rasch model (χ213=43.4; p=0.080; χ212=17.5, p=0.130; χ26=32.9, p=0.040). Their reliability (Person Separation Index=0.912; 0.900; 0.800) was adequate for individual (VAE, VOE1) and group measurement (VOE2). The serial combination with ‘AND’ rule of the scales generated FR serial algorithms, with good diagnostic accuracy, in the prediction of the described outcomes in community-dwelling elderly, also with PD and stroke, based on cutoffs defined using an ‘ad hoc’ method, which considered a higher cost of false negatives compared to false positives (≥1 fall: sensitivity (SE)=62.4%; specificity (SP)=71.0%; diagnostic accuracy (DA)=0.672; ≥2 falls: SE=72.8%; SP=63.2%; DA=0.657; ≥3 falls: SE=79.3%; SP=60.0%; DA=0.629). We calculated cumulative post-test probabilities of the combined scales, which performed more effectively than single tools, and we constructed additional algorithms based on logistic regression models using a parallel combination. We realized an external validation through the comparison with FRAT-up algorithms. Then, we demonstrated the management with Rasch analysis of the neurological disease effect on tools performance (e.g., VAE scale: splitting analysis for DIF by neurological diseases). Finally, we validated an ICD&ICF core set for the FR in community-dwelling elderly, also with PD and stroke (103 FR factors linked to 74 categories). Further projects are desirable to replicate these findings in larger, multicenter validation studies.
Il rischio di caduta (RdC) e una lesione correlata aumentano con l'età e con l'associazione di patologie neurologiche. Le cadute sono un evento temibile per le conseguenze traumatiche e psicologiche, il loro costo sta diventando insostenibile e la loro prevenzione negli anziani è una priorità di sanità pubblica. Le linee guida internazionali raccomandano la rimozione dei fattori di RdC modificabili e l'attuazione di interventi efficaci. Tuttavia, non sono state utilizzate classificazioni univoche dei fattori di RdC, quindi l'uso delle Classificazioni Internazionali dell'OMS come riferimento universale può essere la soluzione più logica e naturale. Lo screening del RdC è la prima componente di programmi di prevenzione delle cadute efficaci. Ad oggi, nonostante l'utilizzo di numerosi strumenti di valutazione, non è possibile rilevare e prevedere il RdC con un’accuratezza diagnostica ottimale. Gli obiettivi della tesi sono stati: 1) validare un algoritmo di screening seriale del RdC con alta accuratezza diagnostica in un campione di anziani in comunità, anche con PD e ictus, nella predizione di una, due e tre cadute nei successivi dodici mesi; 2a) valutare l'effetto delle patologie neurologiche sui test di screening; 2b) validare un core set ICD&ICF per il RdC nella stessa popolazione. Utilizzando i dati raccolti nello studio PRE.C.I.S.A., abbiamo eseguito le seguenti analisi: 1) per validare un algoritmo di screening seriale abbiamo calibrato le scale VAE, VOE1 e VOE2 con tecniche classiche e Rasch analysis, calcolato i due FRAT-up e studiato l’accuratezza diagnostica degli strumenti e degli algoritmi di screening, ottenuti con combinazioni seriali delle scale e dei due FRAT-up e con modelli di regressione logistica, nella predizione degli outcome descritti; abbiamo confrontato le loro proprietà per validazione esterna; 2a) per valutare l'effetto delle patologie neurologiche sui test abbiamo condotto l’analisi del Differential Item Functioning (DIF) per le scale e il t-test per i due FRAT-up; 2b) per validare il core set ICD&ICF abbiamo revisionato i fattori di RdC e linkato alle categorie classificative. Il campione disponibile dallo studio PRE.C.I.S.A. è stato di 768 anziani (femmine 65.3%; età media 75.8). Partendo da 29 variabili, abbiamo calibrato tre scale di misurazione (VAE, VOE1 e VOE), che hanno mostrato un fit soddisfacente al Rasch model (χ213=43.4; p=0.080; χ212=17.5, p=0.130; χ26=32.9, p=0.040). La loro affidabilità (Person Separation Index=0.912; 0.900; 0.800) è stata adeguata per la misurazione individuale (VAE, VOE1) e di gruppo (VOE2). La combinazione seriale con ‘AND’ rule delle scale ha generato algoritmi seriali con buona accuratezza diagnostica nella predizione degli outcome descritti in anziani in comunità, anche con PD e ictus, sulla base di cutoff definiti con un metodo ‘ad hoc’, che ha considerato un costo maggiore dei falsi negativi rispetto ai falsi positivi (1 caduta: sensibilità (SE)=62.4%; specificità (SP)=71.0%; accuratezza diagnostica (AD)=0.672; 2 cadute: SE=72.8%; SP=63.2%; AD=0.657; 3 cadute: SE=79.3%; SP=60.0%; AD=0.629). Abbiamo calcolato le probabilità cumulative post-test delle scale combinate con performance migliori di quelle dei singoli strumenti e costruito algoritmi aggiuntivi, basati su modelli di regressione logistica, utilizzando una combinazione parallela. Abbiamo realizzato una validazione esterna attraverso il confronto con algoritmi costruiti con i FRAT-up. Abbiamo dimostrato, poi, la gestione con Rasch analysis dell'effetto delle malattie neurologiche sulle prestazioni dei test (VAE: splitting analysis per DIF da malattie neurologiche). Infine, abbiamo validato un core set ICD&ICF per il RdC negli anziani in comunità, anche con PD e ictus (103 fattori di RdC linkati a 74 categorie). Sono auspicabili ulteriori studi multicentrici di validazione per replicare questi risultati.
(2021). Fall risk detection and prediction in community-dwelling older adults. (Tesi di dottorato, Università degli Studi di Milano-Bicocca, 2021).
Fall risk detection and prediction in community-dwelling older adults
CASELLI, SERENA
2021
Abstract
The fall risk (FR) and a related injury increase with age and with the association of neurological diseases (Parkinson’s Disease (PD) or stroke). Falls represent a fearsome event for an elderly for traumatic and psychological consequences, and costs are becoming unsustainable. Hence, fall prevention in elderly at risk is a public health priority. All international guidelines recommend removing the modifiable FR factors and implementing effective interventions on people at risk. Nevertheless, investigators have not used consistent classifications for FR factors, so using the WHO Family of International Classifications can be the more natural and logical solution to cover the lack of a universal reference framework. FR screening is the first component of effective fall prevention programs. To date, despite the use of numerous FR assessment tools, it is not possible to detect and predict elderly fallers with optimal diagnostic accuracy. The aims of the thesis were: 1) to validate a FR serial screening algorithm with high diagnostic accuracy in a sample of community-dwelling elderly, also with PD and stroke, in the prediction of at least one, two, and three falls in the following twelve months; 2a) to assess the neurological disease's effect on FR tests; 2b) to validate an ICD& ICF core set for FR in the same population. Using data collected in the PRE.C.I.S.A. trial on fall risk, we performed the following analyses: 1) to validate a FR screening algorithm we calibrated the VAE, VOE1 and VOE2 scales with classical and Rasch analysis and we calculated the two FRAT-up; we studied the diagnostic accuracy of single tools and screening algorithms, obtained with serial combinations of the scales and the two FRAT-up tools, and with logistic regression models, in the prediction of the described outcomes; we compared their properties for external validation; 2a) to assess neurological disease’s effect on the tools we conducted a Differential Item Functioning (DIF) analysis for the scales and a t-test for the two FRAT-up; 2b) to validate the core set we reviewed the FR factors and we linked them to classification categories. The available sample from the PRE.C.I.S.A. trial was constituted by 768 older adults (female 65.3%; mean age 75.8). From 29 observed variables, we calibrated three measurement scales (VAE, VOE1, and VOE), which showed a satisfactory fit to the Rasch model (χ213=43.4; p=0.080; χ212=17.5, p=0.130; χ26=32.9, p=0.040). Their reliability (Person Separation Index=0.912; 0.900; 0.800) was adequate for individual (VAE, VOE1) and group measurement (VOE2). The serial combination with ‘AND’ rule of the scales generated FR serial algorithms, with good diagnostic accuracy, in the prediction of the described outcomes in community-dwelling elderly, also with PD and stroke, based on cutoffs defined using an ‘ad hoc’ method, which considered a higher cost of false negatives compared to false positives (≥1 fall: sensitivity (SE)=62.4%; specificity (SP)=71.0%; diagnostic accuracy (DA)=0.672; ≥2 falls: SE=72.8%; SP=63.2%; DA=0.657; ≥3 falls: SE=79.3%; SP=60.0%; DA=0.629). We calculated cumulative post-test probabilities of the combined scales, which performed more effectively than single tools, and we constructed additional algorithms based on logistic regression models using a parallel combination. We realized an external validation through the comparison with FRAT-up algorithms. Then, we demonstrated the management with Rasch analysis of the neurological disease effect on tools performance (e.g., VAE scale: splitting analysis for DIF by neurological diseases). Finally, we validated an ICD&ICF core set for the FR in community-dwelling elderly, also with PD and stroke (103 FR factors linked to 74 categories). Further projects are desirable to replicate these findings in larger, multicenter validation studies.File | Dimensione | Formato | |
---|---|---|---|
phd_unimib_835780.pdf
Accesso Aperto
Descrizione: Tesi Serena Caselli
Tipologia di allegato:
Doctoral thesis
Dimensione
6.11 MB
Formato
Adobe PDF
|
6.11 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.