Choroby zakaźne
Rokowania, prognozy i postęp choroby

Prognozowanie przebiegu chorób zakaźnych jest kluczowe dla optymalizacji leczenia i alokacji zasobów w opiece zdrowotnej, jednak napotyka na liczne wyzwania, takie jak heterogeniczność chorób, zmienność reakcji pacjentów oraz opóźnienia w raportowaniu przypadków. Tradycyjne metody prognostyczne, oparte na skalach klinicznych i biomarkerach (np. prokalcytonina w sepsie, D-dimer, IL-6, ferrytyna i CRP w COVID-19), są uzupełniane przez zaawansowane modele uczenia maszynowego, takie jak random forest z dokładnością 97% w przewidywaniu ciężkiego przebiegu COVID-19 czy modele DNN i LSTM przewyższające klasyczne ARIMA. Transfer learning, jak w metodologii TransMED, poprawia predykcję u pacjentów z COVID-19 o 12,9% i 10,3% w AUROC, wykorzystując dane z chorób układu oddechowego. Systematyczne przeglądy wykazały jednak, że większość modeli prognostycznych COVID-19 (90% z 593 modeli) cechuje wysokie ryzyko błędu, a tylko nieliczne, takie jak Qcovid, PRIEST i ISARIC 4C Deterioration, mają niskie ryzyko błędu i są rekomendowane do dalszej walidacji.

Wprowadzenie do prognozowania w chorobach zakaźnych

Choroby zakaźne stanowią istotne zagrożenie dla zdrowia publicznego na całym świecie. Dokładne przewidywanie przebiegu i wyniku tych chorób jest kluczowe dla efektywnego leczenia pacjentów oraz optymalnej alokacji zasobów opieki zdrowotnej. Wczesna identyfikacja pacjentów zagrożonych ciężkim przebiegiem choroby może umożliwić bardziej ukierunkowane, terminowe interwencje i skuteczną komunikację ryzyka, co ma wymierny wpływ na rozprzestrzenianie się choroby w scenariuszach epidemicznych i pandemicznych.1

Prognostyka w chorobach zakaźnych napotyka jednak na znaczące wyzwania, wynikające z heterogeniczności tych chorób, zmienności reakcji pacjentów na zakażenie oraz opóźnień w raportowaniu przypadków. Pomimo postępów w dziedzinie modelowania predykcyjnego, nie istnieje pojedynczy złoty standard dla przewidywania przebiegu chorób zakaźnych. Obecne badania eksplorują szeroki zakres narzędzi prognostycznych – od tradycyjnych systemów punktacji i biomarkerów po zaawansowane technologie omiczne i sztuczną inteligencję.2

Metody prognozowania w chorobach zakaźnych

Tradycyjne skale i biomarkery

Tradycyjne metody prognozowania w chorobach zakaźnych opierają się na skalach klinicznych i biomarkerach. W przypadku sepsy, która jest stanem zagrażającym życiu, wymagającym precyzyjnej prognozy dla złagodzenia niekorzystnych wyników, stosuje się różne skale, jak np. Mortality in Emergency Department Sepsis score w połączeniu z prokalcytoniną.3 W grypie czynniki prognostyczne obejmują choroby układu sercowo-naczyniowego i objawy ze strony ośrodkowego układu nerwowego.4

W kontekście COVID-19, badania wykazały, że różne zestawy biomarkerów mogą być stosowane do oceny ciężkości choroby i prognozy. Na przykład, dwa różne zestawy biomarkerów (D-dimer, IL-6 i ferrytyna oraz CRP, D-dimer i IL-6) okazały się przydatne w ocenie ciężkości choroby i przewidywaniu wyników klinicznych u pacjentów z COVID-19 w Wietnamie.56

Zaawansowane modele predykcyjne

Postęp w dziedzinie uczenia maszynowego i analizy danych umożliwił rozwój bardziej zaawansowanych modeli predykcyjnych. Model random forest zastosowany w badaniu wietnamskim mógł przewidzieć z 97% dokładnością prawdopodobieństwo pogorszenia stanu pacjentów z COVID-19 do ciężkiego stanu. Co istotne, model ten zachowywał 92% dokładność nawet po usunięciu IL-6 z analizy, co zwiększa jego zastosowanie w szpitalach o ograniczonych możliwościach testowania IL-6.7

W innym badaniu porównano wydajność głębokiej sieci neuronowej (DNN) i modeli uczenia się typu long-short term memory (LSTM) z modelem autoregresyjnym zintegrowanej średniej ruchomej (ARIMA) w przewidywaniu trzech chorób zakaźnych z tygodniowym wyprzedzeniem. Wyniki pokazały, że modele DNN i LSTM działają lepiej niż ARIMA.8 Badanie to wykorzystywało dane o występowaniu chorób zakaźnych, dane z wyszukiwarek internetowych, dane z mediów społecznościowych Twitter oraz dane pogodowe, takie jak temperatura i wilgotność.9

Transfer Learning i modele wielomodalne

Obiecującym podejściem do budowy modeli predykcyjnych dla chorób o ograniczonej ilości danych jest transfer learning. TransMED, metodologia rozwoju wielomodalnych modeli predykcyjnych, adresuje problemy związane z niedostatkiem danych treningowych dla nowych lub rzadkich chorób poprzez transfer learning z chorób o podobnych charakterystykach na poziomie kohorty i podobnych wynikach.10

Eksperymenty pokazały, że hierarchiczne podejście oparte na transfer learningu wykorzystujące kohorty z ciężkimi chorobami układu oddechowego (SRD) prowadzi do średniej poprawy o 12,9% i 10,3% w AUROC dla przewidywania pobytu i wentylacji pacjentów z COVID-19.11 Kluczowym wnioskiem z tego badania jest to, że uczenie hierarchiczne, które najpierw modeluje interakcję między różnymi pojęciami medycznymi w krótszych odstępach czasu, a następnie uczy się zależności czasowych, jest skuteczne dla transfer learningu między chorobami, w których stan pacjenta ewoluuje w różnych skalach czasowych.12

Prognostyka w specyficznych chorobach zakaźnych

COVID-19

Pandemia COVID-19 ujawniła istotne wyzwania w rozwijaniu systemów, które mogą dokładnie przewidywać wyniki związane z nowo pojawiającą się chorobą zakaźną.13 Takie zdolności są kluczowe dla instytucji, aby priorytetyzować zasoby i wdrażać ilościowe podejście do triażu w sytuacji awaryjnej.14

Systematyczny przegląd i krytyczna ocena modeli prognostycznych COVID-19 wykazały, że dostępne są różnorodne modele predykcyjne, ale ich jakość i użyteczność kliniczna są zmienne. W jednym z przeglądów zidentyfikowano 731 modeli, w tym 125 modeli diagnostycznych i 606 modeli prognostycznych. Większość z tych modeli była jednak słabo raportowana i obarczona wysokim ryzykiem błędu, co sugeruje, że ich rzeczywista wydajność predykcyjna w praktyce jest prawdopodobnie niższa niż raportowana.15

Spośród 593 modeli prognostycznych dla pacjentów z COVID-19, 90% (n=536) było obciążonych wysokim ryzykiem błędu. Tylko siedem nowo opracowanych modeli miało niskie ryzyko błędu, w tym cztery modele Qcovid przewidujące przyjęcie do szpitala i śmierć z powodu COVID-19 w populacji ogólnej w Wielkiej Brytanii, wynik PRIEST przewidujący 30-dniową śmierć lub wsparcie narządów u pacjentów z podejrzeniem lub potwierdzonym COVID-19 zgłaszających się na oddział ratunkowy, oraz model ISARIC 4C Deterioration, który został zwalidowany regionalnie w Wielkiej Brytanii.16

W innym badaniu zidentyfikowano sześć najważniejszych zmiennych dla przewidywania śmiertelności lub przeżycia pacjentów z COVID-19: długość pobytu w szpitalu, czas trwania pobytu na OIT, czas od przyjęcia do OIT, dni do wypisu żywego lub śmierci, D-dimer i pH krwi.17 Inne cechy, takie jak wiek przy przyjęciu, stosunek PaO2/FiO2, TropT, ferrytyna, wentylacja, białko C-reaktywne (CRP) i objawy zespołu ostrej niewydolności oddechowej (ARDS), również są krytycznymi determinantami wyniku.18

W jeszcze innym badaniu opracowano i zaślepiło zwalidowano trzy różne testy oparte na uczeniu maszynowym przewidujące trzy miary progresji do ciężkiej choroby u pacjentów hospitalizowanych z COVID-19: ryzyko progresji do przyjęcia na OIT, intubacji i diagnozy ARDS.19 Przewidywanie ryzyka tych krótkoterminowych, częściej występujących punktów końcowych może być bardziej istotne dla prowadzenia rozmów z pacjentami i członkami rodziny dotyczących agresywności opieki i chęci intubacji, kierowania ograniczonych zasobów szpitalnych do pacjentów, którzy najbardziej ich potrzebują, identyfikacji osób, które najprawdopodobniej skorzystają z określonych interwencji lub zapisania się do badań klinicznych, oraz zapewnienia spokoju ducha pacjentom z kategorii niższego ryzyka.20

Gorączka denga

Gorączka denga zyskała miano szybko rozwijającej się globalnej epidemii, gdyż przenosiciel choroby dostosował się do chłodniejszych krajów, przełamując przekonanie, że denga to choroba występująca wyłącznie w strefach tropikalnych/subtropikalnych. Ta „zakaźna bomba zegarowa” wymaga terminowego i odpowiedniego leczenia, ponieważ wpływa na kluczowe funkcje organizmu, często prowadząc do niewydolności wielu narządów, gdy u pacjentów pojawia się małopłytkowość i krwawienie wewnętrzne, zwiększając chorobowość i śmiertelność.21

Opracowany model przewiduje poziomy zakażenia u pacjenta w oparciu o klasyfikację przedstawioną przez Światową Organizację Zdrowia, tj. gorączkę denga, gorączkę krwotoczną denga i zespół wstrząsu denga, osiągając znacząco wysoką dokładność ponad 90%, wraz z wysokimi wartościami czułości i swoistości.22

Predykcje dla analizy poziomu dengi w proponowanym modelu, w przeciwieństwie do tradycyjnego podejścia, nie opierają się wyłącznie na parametrach takich jak liczba płytek krwi, ciśnienie krwi i krwotoki, ale obejmują również profile wątrobowe i nerkowe, ponieważ gdy prognoza choroby od gorączki denga osiąga wyższe poziomy śmiertelności, często występuje zaangażowanie wielu narządów, szczególnie obciążając funkcje wątroby i nerek pacjenta i prowadząc do niewydolności narządów.23

Prognozowanie przebiegu epidemii gorączki denga w czasie rzeczywistym jest również wyzwaniem. Wymaga to wiedzy na temat konkretnego systemu chorobowego, a także potoku, który może przekształcić surowe dane z systemu nadzoru zdrowia publicznego w skalibrowane prognozy zachorowalności na chorobę.24 Jednak możliwość skutecznego przewidywania rozwoju sytuacji jest utrudniona przez opóźnienia w raportowaniu przypadków. Analizy wykazały, że gdyby nie było opóźnień w raportowaniu, model mógłby dokonywać znacznie dokładniejszych prognoz w prawie wszystkich prowincjach tajlandzkich.25

Wyzwania i ograniczenia w prognozowaniu chorób zakaźnych

Pomimo postępów w dziedzinie modelowania predykcyjnego, prognozowanie przebiegu chorób zakaźnych napotyka na różne wyzwania i ograniczenia:

  • Zmienność pacjentów: Badania wykazały, że dokładność predykcji uczenia maszynowego zmniejsza się wraz ze wzrostem zmienności pacjentów. Nawet gdy znany jest pełny stan systemu (tj. wszystkie zmienne pacjenta są dokładnie zmierzone w funkcji czasu), nasza wiedza o wyniku pacjenta jest ograniczona, gdy występuje zmienność między pacjentami.26
  • Kompromis między wczesną a dokładną prognozą: Klinicyści muszą przewidywać wyniki pacjentów z wysoką dokładnością jak najwcześniej po wystąpieniu choroby. Jednak dokładność można zwiększyć kosztem opóźnionej prognozy. Wyniki badań pokazują, że dla wszystkich badanych modeli równań różniczkowych zwyczajnych (ODE) dokładność predykcji wzrasta wraz z upływem czasu tc użytego do klasyfikacji.27
  • Opóźnienia w raportowaniu: Opóźnienia w raportowaniu przypadków znacząco wpływają na dokładność prognoz w czasie rzeczywistym. Analizy dotyczące gorączki denga w Tajlandii wykazały, że brak opóźnień w raportowaniu prowadziłby do znacznie dokładniejszych prognoz.28
  • Ryzyko błędu w modelach: Systematyczne przeglądy modeli prognostycznych COVID-19 wykazały, że większość opublikowanych badań modelowych była słabo raportowana i obarczona wysokim ryzykiem błędu, co sugeruje, że ich rzeczywista wydajność predykcyjna w praktyce jest prawdopodobnie niższa niż raportowana.29
  • Heterogeniczność wyników: Przegląd systematyczny wykazał, że modele prognostyczne COVID-19 były obciążone niedociągnięciami, takimi jak niska jakość metodologiczna, mała wielkość próby, słabe postępowanie z brakującymi danymi, niepowodzenie w radzeniu sobie z nadmiernym dopasowaniem, definicje COVID-19 oparte na cechach klinicznych, a nie na wyniku laboratoryjnego testu diagnostycznego SARS-CoV-2, oraz jego ciężkości z badaniami wykorzystującymi heterogeniczne wyniki.30

Perspektywy na przyszłość

Przyszłe kierunki badań nad prognozowaniem w chorobach zakaźnych obejmują:

  • Standaryzacja metodologii: Istnieje potrzeba przestrzegania wytycznych raportowania, takich jak TRIPOD (transparent reporting of a multivariable prediction model for individual prognosis or diagnosis), w celu poprawy jakości i porównywalności badań modelowych.31
  • Dalsza walidacja obiecujących modeli: Modele o niskim ryzyku błędu, takie jak Qcovid, PRIEST i ISARIC 4C Deterioration, powinny być dalej walidowane w innych zbiorach danych i środowiskach, najlepiej przez niezależnych badaczy, aby zbadać, które modele utrzymują solidną wydajność w czasie i w różnych środowiskach.32
  • Wielkie, wieloośrodkowe badania prospektywne: Przyszłe duże, wieloośrodkowe i dobrze zaprojektowane badania prospektywne są potrzebne do opracowania modeli predykcyjnych dla COVID-19 o użyteczności klinicznej, które mogą być stosowane w różnych populacjach.33
  • Rozwój systemów wczesnego ostrzegania: Istnieją inicjatywy mające na celu opracowanie systemów wczesnego ostrzegania dla nadchodzących fal lub nawet epidemii chorób zakaźnych, takich jak projekt TeCoMed dla grypy i zapalenia oskrzeli.34
  • Integracja różnorodnych źródeł danych: Przyszłe modele predykcyjne mogą czerpać korzyści z integracji danych z różnych źródeł, w tym z systemów elektronicznej dokumentacji medycznej, mediów społecznościowych, danych wyszukiwania w internecie i danych pogodowych, co wykazało swoją wartość w prognozowaniu chorób zakaźnych.35
  • Opracowanie kompleksowych algorytmów predykcyjnych: Istnieją propozycje opracowania testów, które dokładnie przewidują krótko- i długoterminowe (w ciągu jednego roku) wyniki u hospitalizowanych pacjentów z COVID-19, odzwierciedlających szerokie dane demograficzne USA, którzy są narażeni na zwiększone ryzyko niekorzystnych wyników z powodu COVID-19, wykorzystując zarówno dane kliniczne, jak i molekularne.36

Podsumowanie

Prognozowanie wyników w chorobach zakaźnych jest dynamicznie rozwijającym się polem, które ma ogromny potencjał wpływu na opiekę zdrowotną. Od tradycyjnych skal i biomarkerów po zaawansowane modele uczenia maszynowego i transfer learning, różnorodne podejścia są badane i opracowywane w celu poprawy dokładności prognoz.

Jednak istnieją znaczące wyzwania, w tym zmienność pacjentów, kompromis między wczesną a dokładną prognozą, opóźnienia w raportowaniu i ryzyko błędu w modelach. Przyszłe badania powinny skupić się na standaryzacji metodologii, walidacji obiecujących modeli w różnych środowiskach, prowadzeniu dużych, wieloośrodkowych badań prospektywnych i rozwijaniu systemów wczesnego ostrzegania.

Dokładne i możliwe do wdrożenia prognozy zachorowalności na choroby zakaźne w krótkim i długim czasie poprawią odpowiedź zdrowia publicznego na epidemie, umożliwiając bardziej ukierunkowane, terminowe interwencje i skuteczną komunikację ryzyka, co ma wymierny wpływ na rozprzestrzenianie się chorób w scenariuszach epidemicznych i pandemicznych.37

Kolejne rozdziały

Zapraszamy do dalszego czytania naszego leksykonu.

Wybierz kolejny rozdział z menu poniżej, aby otworzyć nową podstronę kompedium wiedzy i uzyskać szczegółowe informację o leku, substancji lub chorobie.

  1. 16.04.2026
  2. www.leksykon.com.pl

Materiały źródłowe

  • #1 Challenges in Real-Time Prediction of Infectious Disease: A Case Study of Dengue in Thailand | PLOS Neglected Tropical Diseases
    https://journals.plos.org/plosntds/article?id=10.1371/journal.pntd.0004761
    Producing accurate and actionable forecasts of infectious disease incidence at short and long time scales will improve public health response to outbreaks. […] Predicting the course of infectious disease outbreaks in real-time is a challenging task. It requires knowledge of the particular disease system as well as a pipeline that can turn raw data from a public health surveillance system into calibrated predictions of disease incidence. […] Improving real-time predictions can enable more targeted, timely interventions and risk communication, both of which have a measurable impact on disease spread in epidemic and pandemic scenarios. […] Real-time forecasts of infectious disease outbreaks can facilitate targeted intervention and prevention strategies, such as increased healthcare staffing or vector control measures.
  • #2
    https://actamedindones.org/index.php/ijim/article/view/2910
    Sepsis is a critical, life-threatening condition that demands precise prediction to mitigate adverse outcomes. The heterogeneity of sepsis leads to variable prognoses, making early and accurate identification increasingly difficult. Despite ongoing advancements, no single gold standard has emerged for sepsis prediction. Current research explores a range of prognostic tools, from traditional scoring systems and biomarkers to cutting-edge omics technologies and artificial intelligence. […] This review aims to critically evaluate the development and application of outcome prediction modalities in sepsis and other infectious diseases, highlighting the progress made and identifying areas for further research.
  • #3 Prediction models for prognosis of influenza: a systematic review and critical appraisal
    https://www.signavitae.com/articles/10.22514/sv.2021.148
    Outcome prediction using the Mortality in Emergency Department Sepsis score combined with procalcitonin for influenza patients. […] Prognostic factors for fatal adult influenza pneumonia. […] Predicting mortality in hospitalized patients with 2009 H1N1 influenza pneumonia. […] Mortality prediction to hospitalized patients with influenza pneumonia: PO2 /FiO2 combined lymphocyte count is the answer.
  • #4 Prediction models for prognosis of influenza: a systematic review and critical appraisal
    https://www.signavitae.com/articles/10.22514/sv.2021.148
    The influenza epidemic has become an important public health issue throughout the world. Early recognition of potentially terrible outcomes is important in the emergency department (ED). Efficient prognosis of the disease is conducive to reducing the financial burden and providing appropriate care for patients. Prediction models containing several features to estimate the risk of patients with confirmed infection could help clinicians give appropriate treatment when health care resources are limited. […] Cardiovascular disease and central nervous symptoms play an important role in prognostic models of influenza. In addition, some commonly used scoring systems can also play a certain role in evaluation. This systematic review compared different types of models for predicting the prognosis of influenza infection, informing us of risk factors for the predictive model in predicting the prognosis of influenza in the early stage.
  • #5 Outcome prediction model and prognostic biomarkers for COVID-19 patients in Vietnam
    https://pmc.ncbi.nlm.nih.gov/articles/PMC9885243/
    Accurate prognosis is important either after acute infection or during long-term follow-up of patients infected by severe acute respiratory syndrome coronavirus 2. This study aims to predict coronavirus disease 2019 (COVID-19) severity based on clinical and biological indicators, and to identify biomarkers for prognostic assessment. […] The random forest model could predict with 97% accuracy the likelihood of COVID-19 patients who subsequently worsened to the severe condition. […] The model could still predict with 92% accuracy after removing IL-6 from the analysis to generalise the applicability of the model to hospitals with limited capacity for IL-6 testing. […] Two different sets of biomarkers (D-dimer, IL-6 and ferritin, and CRP, D-dimer and IL-6) are applicable for the assessment of disease severity and prognosis.
  • #6 Outcome prediction model and prognostic biomarkers for COVID-19 patients in Vietnam
    https://pmc.ncbi.nlm.nih.gov/articles/PMC9885243/
    This study provided a simple and reliable model using two different sets of biomarkers to assess disease severity and predict clinical outcomes in COVID-19 patients in Vietnam. […] The outcome of SARS-CoV-2 infection is largely dependent on viral virulence and variability as well as the host’s genetic background. […] Although serum levels of D-dimer, IL-6, CRP and ferritin have been well documented as biomarkers for the prediction of disease progression and mortality, and guide the management of COVID-19 patients in different world populations, there is no study describing the prognostic values of combinations of these biomarkers with various clinical parameters in COVID-19 patients living in low- and middle-income countries (LMICs) such as Vietnam. […] Our study results demonstrated that CRP, D-dimer, ferritin and IL-6 were significantly correlated with the prognosis of severe COVID-19 patients. Furthermore, this study provided a simple and reliable model to predict clinical outcomes for COVID-19 patients in Vietnam. Two different sets of biomarkers are applicable for the assessment of disease severity and prognosis.
  • #7 Outcome prediction model and prognostic biomarkers for COVID-19 patients in Vietnam
    https://pmc.ncbi.nlm.nih.gov/articles/PMC9885243/
    Accurate prognosis is important either after acute infection or during long-term follow-up of patients infected by severe acute respiratory syndrome coronavirus 2. This study aims to predict coronavirus disease 2019 (COVID-19) severity based on clinical and biological indicators, and to identify biomarkers for prognostic assessment. […] The random forest model could predict with 97% accuracy the likelihood of COVID-19 patients who subsequently worsened to the severe condition. […] The model could still predict with 92% accuracy after removing IL-6 from the analysis to generalise the applicability of the model to hospitals with limited capacity for IL-6 testing. […] Two different sets of biomarkers (D-dimer, IL-6 and ferritin, and CRP, D-dimer and IL-6) are applicable for the assessment of disease severity and prognosis.
  • #8 Predicting Infectious Disease Using Deep Learning and Big Data
    https://www.mdpi.com/1660-4601/15/8/1596
    Infectious disease occurs when a person is infected by a pathogen from another person or an animal. It is a problem that causes harm at both individual and macro scales. […] The performance of the deep neural network (DNN) and long-short term memory (LSTM) learning models were compared with the autoregressive integrated moving average (ARIMA) when predicting three infectious diseases one week into the future. The results show that the DNN and LSTM models perform better than ARIMA. […] We believe that this study’s models can help eliminate reporting delays in existing surveillance systems and, therefore, minimize costs to society. […] An increasing number of researchers recognize these facts and are performing data-based infectious disease surveillance studies to supplement existing systems and design new models.
  • #9 Predicting Infectious Disease Using Deep Learning and Big Data
    https://www.mdpi.com/1660-4601/15/8/1596
    The aim of this study is to design a model that uses the infectious disease occurrence data provided by the KCDC, search query data from search engines that are specialized for South Korea, Twitter social media big data, and weather data such as temperature and humidity. […] Ultimately, using the results obtained by this study, it should be possible to create a model that can predict trends about the occurrence of infectious disease in real time. Such a model can not only eliminate the reporting time differences in conventional surveillance systems but also minimize the societal costs and economic losses caused by infectious disease. […] The results of OLS analysis using optimal parameters showed that the regression models for each infectious disease had significant results. Of the four input variables, the Naver search frequency had a significant relationship with all three infectious diseases.
  • #10 Preparing for the next pandemic via transfer learning from existing diseases with hierarchical multi-modal BERT: a study on COVID-19 outcome prediction | Scientific Reports
    https://www.nature.com/articles/s41598-022-13072-w
    Such capabilities are critical for institutions to prioritize resources, and bring a quantitative approach to triaging in an emergency, which subjects the human caregivers to intense psychological stress. […] Due to the lack of historical COVID-19 cases for training supervised machine learning models, early methods for COVID-19 severity prediction focused on the analysis of a relatively small number of carefully chosen model covariates, which included demographic risk factors, prior comorbidities, symptoms on admission, and laboratory biomarkers. […] We propose TransMED, a methodology for developing multi-modal predictive models, while addressing training data scarcity issues posed by emerging (or rare) diseases through transfer learning from diseases with shared cohort-level characteristics and similar outcomes.
  • #11 Preparing for the next pandemic via transfer learning from existing diseases with hierarchical multi-modal BERT: a study on COVID-19 outcome prediction | Scientific Reports
    https://www.nature.com/articles/s41598-022-13072-w
    To address the existing gaps in pandemic preparedness, we sought to improve on current methods to: (i) predict if a patient will be staying in the hospital, after a certain time using the patients multi-modal history. […] This provides a better understanding of the severity of a patients condition, and (ii) predict the likelihood of a patient requiring mechanical ventilation. […] Our experiments show that our hierarchical transfer learning based approach using Severe Respiratory Disease (SRD) cohorts leads to an average improvement of 12.9% and 10.3% in AUROC for COVID-19 patient stay and ventilation prediction. […] We believe our analysis would motivate data-driven discovery of key multi-comorbidities associated with a disease while advancing the interpretability and rigor for evaluating deep learning models for clinical use.
  • #12 Preparing for the next pandemic via transfer learning from existing diseases with hierarchical multi-modal BERT: a study on COVID-19 outcome prediction | Scientific Reports
    https://www.nature.com/articles/s41598-022-13072-w
    Our work shows that a transfer-learning approach that learns from prior and related EHR databases is a promising way to build predictive models for diseases with limited data. […] A key conclusion from our study is that hierarchical learning, that first models the interaction between various medical concepts over shorter intervals, and then learns temporal dependencies is effective for transfer learning across diseases where patient conditions evolve at different time-scales.
  • #13 Preparing for the next pandemic via transfer learning from existing diseases with hierarchical multi-modal BERT: a study on COVID-19 outcome prediction | Scientific Reports
    https://www.nature.com/articles/s41598-022-13072-w
    Developing prediction models for emerging infectious diseases from relatively small numbers of cases is a critical need for improving pandemic preparedness. […] We demonstrate the alignment of TransMEDs predictions with well-established clinical knowledge about COVID-19 through univariate and multivariate risk factor driven sub-cohort analysis. […] TransMEDs superior performance over state-of-the-art methods shows that leveraging patient data across modalities and transferring prior knowledge from similar disorders is critical for accurate prediction of patient outcomes, and this approach may serve as an important tool in the early response to future pandemics. […] The COVID-19 pandemic revealed salient challenges in developing systems that can accurately predict outcomes associated with an emerging infectious disease.
  • #14 Preparing for the next pandemic via transfer learning from existing diseases with hierarchical multi-modal BERT: a study on COVID-19 outcome prediction | Scientific Reports
    https://www.nature.com/articles/s41598-022-13072-w
    Such capabilities are critical for institutions to prioritize resources, and bring a quantitative approach to triaging in an emergency, which subjects the human caregivers to intense psychological stress. […] Due to the lack of historical COVID-19 cases for training supervised machine learning models, early methods for COVID-19 severity prediction focused on the analysis of a relatively small number of carefully chosen model covariates, which included demographic risk factors, prior comorbidities, symptoms on admission, and laboratory biomarkers. […] We propose TransMED, a methodology for developing multi-modal predictive models, while addressing training data scarcity issues posed by emerging (or rare) diseases through transfer learning from diseases with shared cohort-level characteristics and similar outcomes.
  • #15 Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal | The BMJ
    https://www.bmj.com/content/369/bmj.m1328
    Objective To review and appraise the validity and usefulness of published and preprint reports of prediction models for prognosis of patients with covid-19, and for detecting people in the general population at increased risk of covid-19 infection or being admitted to hospital or dying with the disease. […] Of these 731, 125 were diagnostic models (including 75 based on medical imaging) and the remaining 606 were prognostic models for either identifying those at risk of covid-19 in the general population (13 models) or predicting diverse outcomes in those individuals with confirmed covid-19 (593 models). […] The reported C indexes varied between 0.77 and 0.93 in development studies with low risk of bias, and between 0.56 and 0.78 in external validations with low risk of bias. […] Most published prediction model studies were poorly reported and at high risk of bias such that their reported predictive performances are probably optimistic.
  • #16 Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal | The BMJ
    https://www.bmj.com/content/369/bmj.m1328
    The 593 prognostic models for patients with covid-19 were more often at high risk of bias than the 13 general population models (90% (n=536) v 69% (n=9)). […] We found seven newly developed models at low risk of bias. […] The four Qcovid models predict hospital admission and death with covid-19 in the general population in the UK, separately for men and women. […] The PRIEST score predicts 30 day death or organ support in patients with suspected or confirmed covid-19 presenting at the emergency department. […] The ISARIC 4C Deterioration model was validated regionally within the UK. […] We suggest that these models should be further validated within other datasets and settings, and ideally by independent investigators, to investigate which models maintain a robust performance over time and in varying settings.
  • #17 Predictive modelling and identification of critical variables of mortality risk in COVID-19 patients | Scientific Reports
    https://www.nature.com/articles/s41598-023-46712-w
    The study aimed to investigate the performance and interpretability of several ML algorithms for predicting COVID-19 mortality risk with an emphasis on the effect of cross-validation (CV) and principal component analysis (PCA) on the results. […] Accurate COVID-19 mortality prediction is essential because it will enable health management systems to allocate adequate and appropriate healthcare resources to the most critical COVID-19 cases. […] The study reveals how ML could assist medical practitioners in making informed decisions on handling critically ill COVID-19 patients with comorbidities. […] The six most critical variables for predicting the mortality or survival of COVID-19 patients were: Length of stay in the hospital, Duration in ICU, Time to ICU from Admission, Days discharged alive or death, D-dimer, and Blood pH.
  • #18 Predictive modelling and identification of critical variables of mortality risk in COVID-19 patients | Scientific Reports
    https://www.nature.com/articles/s41598-023-46712-w
    Other features such as Age at admission, PaO2/FiO2 ratio, TropT, Ferritin, ventilation, C-reactive protein (CRP), and symptoms of acute respiratory distress syndrome (ARDS) are also critical determinants of outcome in terms of survival or death of a patient. […] The performance of deep MLP and SVM can be enhanced through cross-validation to prevent overfitting and improve model generalisation on unseen data. […] The findings have identified some prognostic factors that strongly relate to comorbidities and the prediction of severity and outcomes of critically ill COVID-19 patients.
  • #19 Predicting Prognosis in COVID-19 Patients using Machine Learning and Readily Available Clinical Data | medRxiv
    https://www.medrxiv.org/content/10.1101/2021.01.29.21250762v1.full-text
    Rationale Prognostic tools for aiding in the treatment of hospitalized COVID-19 patients could help improve outcome by identifying patients at higher or lower risk of severe disease. […] The study objective was to develop models to stratify patients by risk of severe outcomes during COVID-19 hospitalization using readily available information at hospital admission. […] Risk of severe outcomes for patients hospitalized with COVID-19 infection can be assessed using machine learning-based models based on attributes routinely collected at hospital admission. […] Using readily accessible EHR-derived data recorded at the time of hospital admission, we developed and blindly validated three different ML-based risk tests predicting three measures of progression to severe disease in patients hospitalized with COVID-19; namely risk of progression to ICU admission, intubation, and ARDS diagnosis.
  • #20 Predicting Prognosis in COVID-19 Patients using Machine Learning and Readily Available Clinical Data | medRxiv
    https://www.medrxiv.org/content/10.1101/2021.01.29.21250762v1.full-text
    We selected disease severity measures of admission to the ICU, development of ARDS, and intubation as testable severe outcomes. […] Risk prediction of these shorter-term, more commonly occurring endpoints could be more relevant to guide discussions with patients and family members concerning aggressiveness of care and desires for intubation, channeling of scarce hospital resources to patients who would need them most, identification of those most likely to benefit from certain interventions or enrollment in clinical trials, and to provide peace of mind for patients at lower risk categories. […] In summary, we have developed and validated a suite of tests able to assess the risk of a poor outcome for patients hospitalized with COVID-19 based on information easily and routinely collected at time of hospital admission.
  • #21 Predicting Infection Positivity, Risk Estimation, and Disease Prognosis in Dengue Infected Patients by ML Expert System
    https://www.mdpi.com/2071-1050/14/20/13490
    Dengue fever has earned the title of a rapidly growing global epidemic since the disease-causing mosquito has adapted to colder countries, breaking the notion of dengue being a tropical/subtropical disease only. This infectious time bomb demands timely and proper treatment as it affects vital body functions, often resulting in multiple organ failures once thrombocytopenia and internal bleeding manifest in the patients, adding to morbidity and mortality. […] The proposed model predicts infection levels in a patient based on the classification provided by the World Health Organization, i.e., dengue fever, dengue hemorrhagic fever, and dengue shock syndrome, acquiring considerably high accuracy of over 90% along with high sensitivity and specificity values. […] The proposed system also aids in predicting dengue infection levels in the patient, which, according to WHO, are classified based on the increase in severity index as dengue fever, dengue hemorrhagic fever, and dengue shock syndrome.
  • #22 Predicting Infection Positivity, Risk Estimation, and Disease Prognosis in Dengue Infected Patients by ML Expert System
    https://www.mdpi.com/2071-1050/14/20/13490
    Dengue fever has earned the title of a rapidly growing global epidemic since the disease-causing mosquito has adapted to colder countries, breaking the notion of dengue being a tropical/subtropical disease only. This infectious time bomb demands timely and proper treatment as it affects vital body functions, often resulting in multiple organ failures once thrombocytopenia and internal bleeding manifest in the patients, adding to morbidity and mortality. […] The proposed model predicts infection levels in a patient based on the classification provided by the World Health Organization, i.e., dengue fever, dengue hemorrhagic fever, and dengue shock syndrome, acquiring considerably high accuracy of over 90% along with high sensitivity and specificity values. […] The proposed system also aids in predicting dengue infection levels in the patient, which, according to WHO, are classified based on the increase in severity index as dengue fever, dengue hemorrhagic fever, and dengue shock syndrome.
  • #23 Predicting Infection Positivity, Risk Estimation, and Disease Prognosis in Dengue Infected Patients by ML Expert System
    https://www.mdpi.com/2071-1050/14/20/13490
    The predictions for dengue level analysis in the proposed model, unlike the traditional approach, does not only rely on parameters such as platelet count, Bp, and hemorrhaging but also further includes liver and renal profiles, as when disease prognosis from dengue fever reaches higher fatality levels, multiple organ involvement is often present, especially burdening the liver and kidney functions of the patient and leading towards organ failure.
  • #24 Challenges in Real-Time Prediction of Infectious Disease: A Case Study of Dengue in Thailand | PLOS Neglected Tropical Diseases
    https://journals.plos.org/plosntds/article?id=10.1371/journal.pntd.0004761
    Producing accurate and actionable forecasts of infectious disease incidence at short and long time scales will improve public health response to outbreaks. […] Predicting the course of infectious disease outbreaks in real-time is a challenging task. It requires knowledge of the particular disease system as well as a pipeline that can turn raw data from a public health surveillance system into calibrated predictions of disease incidence. […] Improving real-time predictions can enable more targeted, timely interventions and risk communication, both of which have a measurable impact on disease spread in epidemic and pandemic scenarios. […] Real-time forecasts of infectious disease outbreaks can facilitate targeted intervention and prevention strategies, such as increased healthcare staffing or vector control measures.
  • #25 Challenges in Real-Time Prediction of Infectious Disease: A Case Study of Dengue in Thailand | PLOS Neglected Tropical Diseases
    https://journals.plos.org/plosntds/article?id=10.1371/journal.pntd.0004761
    Our ability to make effective predictions into the future in a majority of provinces is made difficult by delayed case reporting. […] Our analyses show that if there were no reporting delays, our model would make substantially more accurate predictions in nearly all of the Thai provinces. […] The predictions described in this manuscript were made available to the MOPH typically within two weeks of the data being delivered to the U.S. research team via a PDF report and a private, interactive web application. […] Continued development and refinement of such prediction pipelines, such as that presented here, will enable existing prediction methods to reach their full potential in making an impact on public health decision-making and planning.
  • #26 Outcome Prediction in Mathematical Models of Immune Response to Infection | PLOS One
    https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0135861
    Clinicians need to predict patient outcomes with high accuracy as early as possible after disease inception. […] However, accuracy can be increased at the expense of delayed prognosis. […] We seek to determine the limits of the prediction accuracy of discrete steady-state outcomes of ODEs as a function of patient variability (i.e. random fluctuations in parameter values) using machine learning techniques. […] Thus, we find that the machine learning prediction accuracy decreases with increasing patient variability. […] Our results emphasize that even when the complete state of the system is known (i.e. all patient variables are measured precisely as a function of time), we have limited knowledge of the patient outcome when there is patient-to-patient variability that gives rise to basin overlap. […] Our results also show that for all of the model ODEs studied the prediction accuracy increases as the time tc used for classification increases. […] In our work, we explicitly show that patient-to-patient fluctuations cause a competition between early and accurate outcome prediction.
  • #27 Outcome Prediction in Mathematical Models of Immune Response to Infection | PLOS One
    https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0135861
    Clinicians need to predict patient outcomes with high accuracy as early as possible after disease inception. […] However, accuracy can be increased at the expense of delayed prognosis. […] We seek to determine the limits of the prediction accuracy of discrete steady-state outcomes of ODEs as a function of patient variability (i.e. random fluctuations in parameter values) using machine learning techniques. […] Thus, we find that the machine learning prediction accuracy decreases with increasing patient variability. […] Our results emphasize that even when the complete state of the system is known (i.e. all patient variables are measured precisely as a function of time), we have limited knowledge of the patient outcome when there is patient-to-patient variability that gives rise to basin overlap. […] Our results also show that for all of the model ODEs studied the prediction accuracy increases as the time tc used for classification increases. […] In our work, we explicitly show that patient-to-patient fluctuations cause a competition between early and accurate outcome prediction.
  • #28 Challenges in Real-Time Prediction of Infectious Disease: A Case Study of Dengue in Thailand | PLOS Neglected Tropical Diseases
    https://journals.plos.org/plosntds/article?id=10.1371/journal.pntd.0004761
    Our ability to make effective predictions into the future in a majority of provinces is made difficult by delayed case reporting. […] Our analyses show that if there were no reporting delays, our model would make substantially more accurate predictions in nearly all of the Thai provinces. […] The predictions described in this manuscript were made available to the MOPH typically within two weeks of the data being delivered to the U.S. research team via a PDF report and a private, interactive web application. […] Continued development and refinement of such prediction pipelines, such as that presented here, will enable existing prediction methods to reach their full potential in making an impact on public health decision-making and planning.
  • #29 Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal | The BMJ
    https://www.bmj.com/content/369/bmj.m1328
    Objective To review and appraise the validity and usefulness of published and preprint reports of prediction models for prognosis of patients with covid-19, and for detecting people in the general population at increased risk of covid-19 infection or being admitted to hospital or dying with the disease. […] Of these 731, 125 were diagnostic models (including 75 based on medical imaging) and the remaining 606 were prognostic models for either identifying those at risk of covid-19 in the general population (13 models) or predicting diverse outcomes in those individuals with confirmed covid-19 (593 models). […] The reported C indexes varied between 0.77 and 0.93 in development studies with low risk of bias, and between 0.56 and 0.78 in external validations with low risk of bias. […] Most published prediction model studies were poorly reported and at high risk of bias such that their reported predictive performances are probably optimistic.
  • #30
    https://link.springer.com/article/10.1007/s10654-023-00973-x
    Model performances were assessed with AUC ranging from 0.57 to 0.99 in 60 studies, and sensitivity and specificity ranging from 7.1 to 100% and 19.5% 100%, respectively. […] Overall, the RoB assessment was rated to be at high or unclear RoB in 312 studies. […] This could be explained by shortcomings such as poor methodological quality, small sample size, poor handling of missing data, failure to deal with overfitting, definitions of COVID-19 based on clinical features rather than on the result of laboratory diagnostic test for SARS-CoV-2 and its severity with studies using heterogeneous outcomes. […] Future large, multi-center and well-designed prospective studies are needed for the development of predictive models for COVID-19 with clinical utility that can be applied to diverse populations.
  • #31 Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal | The BMJ
    https://www.bmj.com/content/369/bmj.m1328
    Finally, prediction modellers should adhere to the TRIPOD (transparent reporting of a multivariable prediction model for individual prognosis or diagnosis) reporting guideline. […] We identified 13 newly developed models aiming to predict covid-19 related risks in the general population. […] We identified 593 prognostic models for predicting clinical outcomes in patients with covid-19 (368 developments, 225 external validations). […] The studies reported C indexes between 0.49 and 1, with a median of 0.81 (interquartile range 0.75-0.89). […] Most newly developed models and external validations were at unclear (n=32, 5%) or high (n=545, 90%) risk of bias according to assessment with PROBAST, which suggests that the predictive performance when used in practice is probably lower than what is reported.
  • #32 Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal | The BMJ
    https://www.bmj.com/content/369/bmj.m1328
    The 593 prognostic models for patients with covid-19 were more often at high risk of bias than the 13 general population models (90% (n=536) v 69% (n=9)). […] We found seven newly developed models at low risk of bias. […] The four Qcovid models predict hospital admission and death with covid-19 in the general population in the UK, separately for men and women. […] The PRIEST score predicts 30 day death or organ support in patients with suspected or confirmed covid-19 presenting at the emergency department. […] The ISARIC 4C Deterioration model was validated regionally within the UK. […] We suggest that these models should be further validated within other datasets and settings, and ideally by independent investigators, to investigate which models maintain a robust performance over time and in varying settings.
  • #33
    https://link.springer.com/article/10.1007/s10654-023-00973-x
    Model performances were assessed with AUC ranging from 0.57 to 0.99 in 60 studies, and sensitivity and specificity ranging from 7.1 to 100% and 19.5% 100%, respectively. […] Overall, the RoB assessment was rated to be at high or unclear RoB in 312 studies. […] This could be explained by shortcomings such as poor methodological quality, small sample size, poor handling of missing data, failure to deal with overfitting, definitions of COVID-19 based on clinical features rather than on the result of laboratory diagnostic test for SARS-CoV-2 and its severity with studies using heterogeneous outcomes. […] Future large, multi-center and well-designed prospective studies are needed for the development of predictive models for COVID-19 with clinical utility that can be applied to diverse populations.
  • #34 Prognosis of Approaching Infectious Diseases | SpringerLink
    https://link.springer.com/chapter/10.1007/978-3-540-39907-0_4
    Few years ago, we have developed an early warning system concerning multiparametric kidney function courses. […] The goal of the TeCoMed project is to compute early warnings against forthcoming waves or even epidemics of infectious diseases in the German federal state of Mecklenburg-Western Pomerania. […] We have developed a prognostic model for diseases that are characterised by cyclic, but irregular behaviour. So far, we have applied this model to influenza and bronchitis.
  • #35 Predicting Infectious Disease Using Deep Learning and Big Data
    https://www.mdpi.com/1660-4601/15/8/1596
    The aim of this study is to design a model that uses the infectious disease occurrence data provided by the KCDC, search query data from search engines that are specialized for South Korea, Twitter social media big data, and weather data such as temperature and humidity. […] Ultimately, using the results obtained by this study, it should be possible to create a model that can predict trends about the occurrence of infectious disease in real time. Such a model can not only eliminate the reporting time differences in conventional surveillance systems but also minimize the societal costs and economic losses caused by infectious disease. […] The results of OLS analysis using optimal parameters showed that the regression models for each infectious disease had significant results. Of the four input variables, the Naver search frequency had a significant relationship with all three infectious diseases.
  • #36 Infectious Diseases UCLA Clinical Trial | COVID-19 Outcome Prediction Algorithm | UCLA Health Clinical Trials and Research Studies
    https://www.uclahealth.org/clinical-trials/covid-19-outcome-prediction-algorithm
    Severe acute respiratory syndrome coronavirus 2-mediated coronavirus disease (COVID-19) is an evolutionarily unprecedented natural experiment that causes major changes to the host immune system. […] We propose to develop a test that accurately predicts short- and long-term (within one-year) outcomes in hospitalized COVID-19 patients broadly reflecting US demographics who are at increased risk of adverse outcomes from COVID-19 using both clinical and molecular data.
  • #37 Challenges in Real-Time Prediction of Infectious Disease: A Case Study of Dengue in Thailand | PLOS Neglected Tropical Diseases
    https://journals.plos.org/plosntds/article?id=10.1371/journal.pntd.0004761
    Producing accurate and actionable forecasts of infectious disease incidence at short and long time scales will improve public health response to outbreaks. […] Predicting the course of infectious disease outbreaks in real-time is a challenging task. It requires knowledge of the particular disease system as well as a pipeline that can turn raw data from a public health surveillance system into calibrated predictions of disease incidence. […] Improving real-time predictions can enable more targeted, timely interventions and risk communication, both of which have a measurable impact on disease spread in epidemic and pandemic scenarios. […] Real-time forecasts of infectious disease outbreaks can facilitate targeted intervention and prevention strategies, such as increased healthcare staffing or vector control measures.