Nowotwór – Rokowania, prognozy i postęp choroby

Nowotwór
Rokowania, prognozy i postęp choroby

Nowotwory stanowią istotną przyczynę chorobowości i śmiertelności globalnie, a precyzyjne prognozowanie przebiegu choroby jest kluczowe dla personalizacji terapii onkologicznej. Tradycyjne metody oparte na parametrach kliniczno-patologicznych, takich jak klasyfikacja TNM, grading, stan węzłów chłonnych, obecność przerzutów oraz ocena sprawności według skali WHO-PS (ECOG-PS), mają ograniczoną skuteczność prognostyczną. W ostatnich latach rozwój sztucznej inteligencji (AI), uczenia maszynowego (ML) i głębokiego uczenia (DL) umożliwił integrację danych klinicznych, genomowych, radiomicznych i histopatologicznych, co znacząco poprawia dokładność przewidywania przeżycia i odpowiedzi na leczenie. Modele DL, takie jak DNN, CNN, DBN, DRN oraz transformatory wizyjne (ViT), wykazują przewagę nad tradycyjnymi metodami, umożliwiając m.in. ocenę ryzyka nawrotu, wczesną diagnostykę oraz dobór optymalnej terapii. Przykłady biomarkerów molekularnych o znaczeniu prognostycznym to 70-genowa sygnatura w raku piersi oraz test Oncotype DX (21 genów), które pomagają w stratyfikacji ryzyka i decyzjach terapeutycznych, choć ich skuteczność może być ograniczona czasowo (np. do 5 lat w przypadku Oncotype DX). Obciążenie mutacyjne guza (TMB) jest istotnym wskaźnikiem u pacjentów poddawanych immunoterapii.

Prognoza nowotworu (predykcja wyników)

Tradycyjne metody prognozowania nowotworów
Zaawansowane metody prognozowania z wykorzystaniem AI i ML
Głębokie uczenie w prognozowaniu nowotworów
Biomarkery i sygnatury molekularne w prognozowaniu
Integracja danych wielomodalnych w prognozowaniu
Specyficzne modele prognostyczne dla poszczególnych typów nowotworów
Nowe podejścia w prognozowaniu nowotworów
Wyzwania i przyszłe kierunki

Podsumowanie

Kolejne rozdziały

Prognoza nowotworu (predykcja wyników)

Nowotwór jest jedną z głównych przyczyn chorobowości i śmiertelności na całym świecie. Prognozowanie przebiegu choroby nowotworowej stanowi kluczowy element opieki onkologicznej, umożliwiający personalizację leczenia i poprawę wyników terapii. Szacowanie prognozy opiera się na danych statystycznych zgromadzonych przez lata badań pacjentów z tym samym typem nowotworu.¹ Tradycyjne metody oceny rokowania pacjentów onkologicznych, oparte wyłącznie na pojedynczych parametrach kliniczno-patologicznych, wykazują ograniczoną skuteczność w przewidywaniu prawdopodobieństwa przeżycia lub odpowiedzi na leczenie.² W ostatnich latach obserwuje się znaczący rozwój zaawansowanych metod predykcyjnych, w tym wykorzystujących sztuczną inteligencję (AI) i uczenie maszynowe (ML), które obiecują dokładniejsze prognozowanie wyników u pacjentów onkologicznych.

Tradycyjne metody prognozowania nowotworów

Standardowo, prognoza w chorobie nowotworowej opiera się na ocenie parametrów klinicznych, takich jak zaawansowanie choroby wg klasyfikacji TNM, stopień zróżnicowania guza (grading), stan węzłów chłonnych czy obecność przerzutów odległych. Dla przykładu, w raku piersi, powszechnie stosowane wytyczne St. Gallen czy kryteria National Institutes of Health (NIH) wykorzystują kombinację tych czynników do stratyfikacji ryzyka.³ System klasyfikacji TNM, uwzględniający trzy istotne czynniki: guz pierwotny (T), zajęcie węzłów chłonnych (N) i przerzuty odległe (M), stanowi podstawę większości modeli prognostycznych, jednak sam w sobie nie uwzględnia indywidualnych cech biologicznych nowotworu.⁴

Ważnym klinicznym parametrem prognostycznym jest również stan sprawności pacjenta oceniany według skali WHO-PS (Eastern Cooperative Oncology Group Performance Status Scale, ECOG-PS), który służy do oceny funkcjonalnego statusu pacjentów onkologicznych na podstawie manifestacji klinicznych guza, aktywności pacjenta i proporcji czasu spędzanego w łóżku.⁵ Chociaż te czynniki kliniczne mają istotne znaczenie prognostyczne, często nie są wystarczające do precyzyjnego określenia rokowania.

Zaawansowane metody prognozowania z wykorzystaniem AI i ML

Sztuczna inteligencja i uczenie maszynowe znacząco zmieniają podejście do prognozowania nowotworów. Metody ML, takie jak Sztuczne Sieci Neuronowe (ANN), Maszyny Wektorów Nośnych (SVM) i Drzewa Decyzyjne (DT), okazały się wysoce skuteczne w przewidywaniu różnych typów nowotworów, w tym raka piersi, mózgu, płuc, wątroby i prostaty.⁶ Podejścia te wykazały większą dokładność w przewidywaniu przebiegu choroby nowotworowej niż tradycyjne metody stosowane przez klinicystów.⁷

Aplikacje AI w onkologii obejmują:⁸

Ocenę ryzyka zachorowania
Wczesną diagnostykę
Szacowanie rokowania pacjenta
Dobór optymalnego leczenia w oparciu o pogłębioną analizę danych

⁹

Algorytmy oparte na sztucznej inteligencji wykazały zdolność do analizy nieustrukturyzowanych danych i precyzyjnego szacowania prawdopodobieństwa zachorowania na różne choroby, w tym nowotwory. Modele AI mogą przewidywać ryzyko nawrotu choroby po zastosowaniu określonej opcji terapeutycznej, co jest kluczowe dla optymalizacji planów dalszego postępowania klinicznego.¹⁰

Głębokie uczenie w prognozowaniu nowotworów

Głębokie uczenie (Deep Learning, DL) jako ważna gałąź sztucznej inteligencji i uczenia maszynowego, znalazło szerokie zastosowanie w różnych aspektach pomocniczej diagnostyki nowotworów, w tym w prognozowaniu ich przebiegu. Modele głębokiego uczenia mogą znacząco poprawić dokładność przewidywania w porównaniu z innymi metodami.¹¹

Główne architektury DL stosowane w prognozowaniu nowotworów obejmują:¹²

Głębokie sieci neuronowe (DNN) – skutecznie integrują dane wieloomiczne, w tym dane kliniczne, patologiczne i radiomiczne
Konwolucyjne sieci neuronowe (CNN) – umożliwiają prognozowanie przeżycia na podstawie obrazów, które może być zintegrowane z istniejącymi danymi klinicznymi
Głębokie sieci przekonań (DBN) – oferują elastyczność dzięki połączeniu uczenia nienadzorowanego i nadzorowanego
Głębokie sieci rekurencyjne (DRN) – szeroko stosowane w analizie trójwymiarowych obrazów medycznych
Transformatory wizyjne (ViT) – stanowią nowe podejście do budowy modeli prognostycznych dla nowotworów

¹³

Badania wykazały, że modele DL oparte na wyżej wymienionych architekturach mogą lepiej integrować różnorodne dane i oferować wyższą dokładność niż tradycyjne modele uczenia maszynowego w przewidywaniu przeżycia pacjentów z nowotworem.¹⁴

Biomarkery i sygnatury molekularne w prognozowaniu

Postęp w dziedzinie genomiki i proteomiki umożliwił identyfikację biomarkerów i sygnatur molekularnych o znaczeniu prognostycznym. Jednym z najbardziej znanych przykładów jest 70-genowa sygnatura w raku piersi opisana przez van’t Veera i współpracowników, która wykazała się wysoką mocą predykcyjną w identyfikacji pacjentów z niskim ryzykiem przerzutów.¹⁵

Profil 70-genowy skutecznie identyfikuje:¹⁶

Pacjentów wysokiego ryzyka, którzy wymagają chemioterapii
Pacjentów niskiego ryzyka, którzy mogliby zrezygnować z chemioterapii

¹⁷

Innym przykładem jest test Oncotype DX oparty na 21 genach, który dostarcza ważnych informacji uzupełniających tradycyjne podejścia patologiczne. Należy jednak zauważyć, że test ten, choć prognostyczny i predykcyjny przed upływem 5 lat, traci te właściwości po tym okresie.¹⁸

Obciążenie mutacyjne guza (Tumor Mutational Burden, TMB), definiowane jako liczba mutacji niesynonimicznych, jest istotnym biomarkerem prognostycznym, szczególnie u pacjentów poddawanych immunoterapii.¹⁹ Wzrastająca liczba dowodów wskazuje, że biomarkery molekularne mogą znacząco poprawić predykcję rokowania.²⁰

Integracja danych wielomodalnych w prognozowaniu

Cyfryzacja dokumentacji medycznej i rosnąca dostępność sekwencjonowania DNA guza stwarzają bezprecedensową możliwość badania determinant wyników leczenia nowotworów. Modele łączące bardziej szczegółowe dane kliniczne, genomiczne, radiomiczne i histopatologiczne wykazały obiecujące wyniki w lepszej stratyfikacji ryzyka.²¹

Badania wykazały, że modele integrujące cechy uzyskane z przetwarzania języka naturalnego (NLP), takie jak lokalizacja choroby, przewyższają te oparte wyłącznie na danych genomowych lub stadium zaawansowania.²² Wyniki wskazują, że biomarkery wielomodalne są lepsze od stadium zaawansowania choroby w prognozowaniu.²³

W niektórych typach nowotworów, dane dotyczące mikrobioty guza mogą również przyczyniać się do dokładniejszej prognozy. Wykazano, że w czterech typach nowotworów: raku nadnercza, raku płaskonabłonkowym szyjki macicy, glejakach niższego stopnia i czerniaku podskórnym, mikrobiota guza jest lepszym predyktorem prognozy niż same zmienne kliniczne.²⁴ Badania wskazują również, że ekspresja genów jest silniejszym predyktorem prognozy w szerszym zakresie typów nowotworów niż ilość mikrobioty.²⁵

Istnieją badania dowodzące, że integracja informacji klinicznych wyodrębnionych przez duże modele językowe (LLM) z analizą obrazów może dodatkowo poprawić dokładność przewidywania przeżycia.²⁶ Badania te wskazują, że LLM mogą dokładnie wyodrębniać informacje kliniczne z dokumentacji medycznej pacjenta, a połączenie automatycznie wyodrębnionych cech klinicznych i obrazowych poprawia skuteczność modeli prognostycznych, np. dla pacjentów z rakiem pęcherza moczowego.²⁷

Specyficzne modele prognostyczne dla poszczególnych typów nowotworów

Dla różnych typów nowotworów opracowano specyficzne modele prognostyczne. Na przykład model PREDICT dla raka piersi to narzędzie online, które pomaga pacjentom i klinicystom ocenić, jak różne terapie mogą poprawić wskaźniki przeżycia po operacji.²⁸ Badanie oceniające wydajność przewidywania PREDICT v2.1 w kohorcie pacjentów z rakiem piersi o wczesnym początku (EoBC) wykazało tendencję do przeszacowywania śmiertelności 5-letniej u osób z przewidywanym ryzykiem ≥30% i śmiertelności 10-letniej u osób z przewidywanym ryzykiem ≥50%.²⁹

W raku szyjki macicy opracowano model predykcyjny przeżycia oparty na miRNA, który stratyfikuje pacjentki na trzy grupy: o wysokim wskaźniku przeżycia (5-letni wskaźnik przeżycia ≥90%), umiarkowanym wskaźniku przeżycia (5-letni wskaźnik przeżycia ~65%) i niskim wskaźniku przeżycia (5-letni wskaźnik przeżycia ≤40%).³⁰ Model ten, opracowany przy użyciu programu SVM (Support Vector Machine), znacząco poprawia użyteczność poprzez bardziej zróżnicowaną stratyfikację pacjentów.³¹

W przewlekłej białaczce limfocytowej (CLL) wykazano, że delecja 11q – a nie same mutacje ATM – jest niezależnym predyktorem wcześniejszej potrzeby leczenia.³² Odkrycie to wzmacnia rolę delecji 11q jako krytycznego markera wyższego ryzyka progresji u pacjentów z CLL, co może pomóc lekarzom w lepszej stratyfikacji ryzyka i personalizacji decyzji terapeutycznych.³³

Nowe podejścia w prognozowaniu nowotworów

Innowacyjne podejścia do prognozowania nowotworów obejmują wykorzystanie cyfrowych zdjęć pacjentów. Badania wykazały, że przy użyciu embedowania zdjęć w przestrzeni ukrytej StyleGAN i nowoczesnych modeli analizy przeżycia, można osiągnąć wskaźnik C wynoszący 0,677, co jest znacznie wyższe niż przypadkowe przewidywanie i świadczy o wartości prognostycznej zawartej w prostych dwuwymiarowych obrazach twarzy.³⁴

Inne podejście, znane jako „Google Goes Cancer”, wykorzystuje metodę rangowania genów według ich znaczenia prognostycznego przy użyciu zarówno informacji o ekspresji, jak i sieci, w sposób podobny do PageRank Google. W badaniu z udziałem 30 pacjentów z rakiem trzustki zidentyfikowano siedem potencjalnych genów markerowych prognostycznych dla wyników leczenia. Sygnatury oparte na tych markerach były niezależnie predykcyjne dla wyniku i lepsze od ustalonych klinicznych czynników prognostycznych, takich jak stopień zróżnicowania, wielkość guza i status węzłów chłonnych.³⁵ Dodatkowa wartość predykcyjna markerów w porównaniu do parametrów klinicznych wynosiła 9% dla pacjentów z terapią adjuwantową i 6% dla pacjentów bez terapii adjuwantowej.³⁶

Wyzwania i przyszłe kierunki

Pomimo znaczących postępów w rozwoju modeli prognostycznych, istnieje wiele wyzwań, które należy pokonać. Wszystkie modele mają wysokie ryzyko błędu systematycznego, co może ograniczać ich zastosowanie w praktyce klinicznej.³⁷ Przyszłe badania powinny koncentrować się na aktualizacji istniejących modeli prognostycznych poprzez dostosowanie predyktorów w celu poprawy ich wydajności oraz promowaniu ich zastosowania klinicznego poprzez zewnętrzną walidację.³⁸

Chociaż różnorodność mutacji nowotworowych stanowi duże wyzwanie dla badaczy opracowujących skuteczne biomarkery dla diagnostyki lub prognozy z powodu dużych proporcji genomu, które muszą być badane w celu zapewnienia odpowiedniej czułości, trwają intensywne prace nad rozwiązaniem tego problemu.³⁹

Najbardziej podstawową rolą klinicznego modelu prognostycznego jest umożliwienie klinicystom i pacjentom zrozumienia rokowania w sposób prosty, szybki i skuteczny, co pozwala na odpowiednie ukierunkowanie leczenia.⁴⁰ Dlatego też, pomimo coraz większej liczby potencjalnych biomarkerów dostarczanych przez technologie sekwencjonowania nowej generacji, istnieje pilna potrzeba uproszczenia wymaganych biomarkerów przy jednoczesnym zapewnieniu stabilności modeli prognostycznych.⁴¹

Podsumowanie

Predykcja wyników w chorobie nowotworowej przechodzi obecnie dynamiczną transformację dzięki zastosowaniu zaawansowanych technologii, takich jak sztuczna inteligencja, uczenie maszynowe i głębokie uczenie. Modele integrujące różnorodne dane – od parametrów klinicznych, przez dane genomiczne, po informacje obrazowe – oferują dokładniejsze przewidywanie przeżycia i odpowiedzi na leczenie niż tradycyjne metody. Identyfikacja i implementacja wiarygodnych biomarkerów prognostycznych, wraz z rozwojem klinicznie efektywnych terapii, jest pilnie potrzebna dla poprawy opieki onkologicznej. Należy jednak pamiętać, że wszystkie modele prognostyczne mają swoje ograniczenia i powinny być stosowane jako narzędzia wspomagające decyzje kliniczne, a nie jako jedyne kryterium wyboru terapii. Dalsze badania nad poprawą istniejących modeli i rozwojem nowych biomarkerów są niezbędne dla postępu medycyny precyzyjnej w onkologii.

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.

Materiały źródłowe

#1 Cancer Prognosis – NCI
https://www.cancer.gov/about-cancer/diagnosis-staging/prognosis
If you have cancer, you may have questions about how serious your cancer is and your chances of survival. The estimate of how the disease will go for you is called prognosis. […] Many people want to know their prognosis. They find it easier to cope when they know more about their cancer. […] Doctors estimate prognosis by using statistics that researchers have collected over many years about people with the same type of cancer. […] Your doctor may tell you that you have a good prognosis if statistics suggest that your cancer is likely to respond well to treatment. Or they may tell you that you have a poor prognosis if the cancer is harder to control. […] If you decide not to have treatment, the doctor who knows your situation best is in the best position to discuss your prognosis. […] Survival statistics most often come from studies that compare treatments with each other, rather than treatment with no treatment. So, it may not be easy for your doctor to give you an accurate prognosis.
#2
https://link.springer.com/article/10.1007/s13167-010-0044-z
Breast cancer is a complex disease, whose heterogeneity is increasingly recognized. […] To date, single clinicopathological parameters show limited success in predicting the likelihood of survival or response to endocrine therapy and chemotherapy. […] Consequently, new gene expression based prognostic and predictive tests are emerging that promise an improvement in predicting survival and therapy response. […] The identification of reliable prognostic biomarkers together with the development of clinically efficient therapies is urgently needed. […] Today, the prognostic clustering of breast cancer in daily routine relies on the determination of a limited set of molecular markers. […] The best HER2-targeted treatment option together with chemotherapy in patients with metastasized but operable breast cancer is currently assessed in clinical trials.
#3
https://link.springer.com/article/10.1007/s13167-010-0044-z
The identification of pattern-based biomarkers for prognosis is a major field of current clinical research in many cancer types including breast cancer. […] Consequently, the identification of prognostic markers identifying the subset of patients eligible for a watchful waiting procedure and/or adjuvant anti-hormonal/anti-HER2 therapy could help to minimize therapy-induced side effects. […] The first breast cancer prognostic signature to be described has been a 70-gene signature by van t Veer et al. […] The 70-gene prognosis profile was a strong predictor of the development of distant metastasis in patients with both lymph-node-negative as well as lymph-node-positive disease. […] The authors then compared the probability that patients classified according to either the 70-gene expression profile, to St. Gallen criteria, or to the National Institutes of Health (NIH) consensus criteria would remain free of distant metastasis.
#4 A narrative review of prognosis prediction models for non-small cell lung cancer: what kind of predictors should be selected and how to improve models?
https://atm.amegroups.org/article/view/81325/html
Compared to the TNM staging system, a prognosis prediction model can improve the accuracy of and guide personalized therapy through the combination of multiple prognostic factors. […] The TNM staging system involves the combination of three important factors, and most prognostic models would take this system into consideration when screening for predictors. […] WHO-PS is the Eastern Cooperative Oncology Group Performance Status Scale (ECOG-PS), which is mainly used to assess the functional status of cancer patients from the clinical manifestation of tumors, patient activity, and the proportion of time in bed. […] The pathological classification of lung cancer has been increasingly detailed with the development and popularization of immunotherapy and targeted therapy. […] Genomics has played a significant part in the prognosis and treatment management of patients with NSCLC by clarifying the role of driver genes and providing information on mutation and gene expression information.
#5 A narrative review of prognosis prediction models for non-small cell lung cancer: what kind of predictors should be selected and how to improve models?
https://atm.amegroups.org/article/view/81325/html
Compared to the TNM staging system, a prognosis prediction model can improve the accuracy of and guide personalized therapy through the combination of multiple prognostic factors. […] The TNM staging system involves the combination of three important factors, and most prognostic models would take this system into consideration when screening for predictors. […] WHO-PS is the Eastern Cooperative Oncology Group Performance Status Scale (ECOG-PS), which is mainly used to assess the functional status of cancer patients from the clinical manifestation of tumors, patient activity, and the proportion of time in bed. […] The pathological classification of lung cancer has been increasingly detailed with the development and popularization of immunotherapy and targeted therapy. […] Genomics has played a significant part in the prognosis and treatment management of patients with NSCLC by clarifying the role of driver genes and providing information on mutation and gene expression information.
#6 Machine Learning and AI in Cancer Prognosis, Prediction, and Treatment Selection: A Critical Approach
https://pmc.ncbi.nlm.nih.gov/articles/PMC10312208/
Cancer is a leading cause of morbidity and mortality worldwide. […] Artificial intelligence (AI), which is used to predict and automate many cancers, has emerged as a promising option for improving healthcare accuracy and patient outcomes. […] AI applications in oncology include risk assessment, early diagnosis, patient prognosis estimation, and treatment selection based on deep knowledge. […] Machine learning (ML), a subset of AI that enables computers to learn from training data, has been highly effective at predicting various types of cancer, including breast, brain, lung, liver, and prostate cancer. […] In fact, AI and ML have demonstrated greater accuracy in predicting cancer than clinicians. […] Therefore, it is important to improve current AI and ML technologies and to develop new programs to benefit patients.
#7 Machine Learning and AI in Cancer Prognosis, Prediction, and Treatment Selection: A Critical Approach
https://pmc.ncbi.nlm.nih.gov/articles/PMC10312208/
Cancer is a leading cause of morbidity and mortality worldwide. […] Artificial intelligence (AI), which is used to predict and automate many cancers, has emerged as a promising option for improving healthcare accuracy and patient outcomes. […] AI applications in oncology include risk assessment, early diagnosis, patient prognosis estimation, and treatment selection based on deep knowledge. […] Machine learning (ML), a subset of AI that enables computers to learn from training data, has been highly effective at predicting various types of cancer, including breast, brain, lung, liver, and prostate cancer. […] In fact, AI and ML have demonstrated greater accuracy in predicting cancer than clinicians. […] Therefore, it is important to improve current AI and ML technologies and to develop new programs to benefit patients.
#8 Machine Learning and AI in Cancer Prognosis, Prediction, and Treatment Selection: A Critical Approach
https://pmc.ncbi.nlm.nih.gov/articles/PMC10312208/
Cancer is a leading cause of morbidity and mortality worldwide. […] Artificial intelligence (AI), which is used to predict and automate many cancers, has emerged as a promising option for improving healthcare accuracy and patient outcomes. […] AI applications in oncology include risk assessment, early diagnosis, patient prognosis estimation, and treatment selection based on deep knowledge. […] Machine learning (ML), a subset of AI that enables computers to learn from training data, has been highly effective at predicting various types of cancer, including breast, brain, lung, liver, and prostate cancer. […] In fact, AI and ML have demonstrated greater accuracy in predicting cancer than clinicians. […] Therefore, it is important to improve current AI and ML technologies and to develop new programs to benefit patients.
#9 Machine Learning and AI in Cancer Prognosis, Prediction, and Treatment Selection: A Critical Approach
https://pmc.ncbi.nlm.nih.gov/articles/PMC10312208/
Cancer is a leading cause of morbidity and mortality worldwide. […] Artificial intelligence (AI), which is used to predict and automate many cancers, has emerged as a promising option for improving healthcare accuracy and patient outcomes. […] AI applications in oncology include risk assessment, early diagnosis, patient prognosis estimation, and treatment selection based on deep knowledge. […] Machine learning (ML), a subset of AI that enables computers to learn from training data, has been highly effective at predicting various types of cancer, including breast, brain, lung, liver, and prostate cancer. […] In fact, AI and ML have demonstrated greater accuracy in predicting cancer than clinicians. […] Therefore, it is important to improve current AI and ML technologies and to develop new programs to benefit patients.
#10 Machine Learning and AI in Cancer Prognosis, Prediction, and Treatment Selection: A Critical Approach
https://pmc.ncbi.nlm.nih.gov/articles/PMC10312208/
Algorithms based on artificial intelligence have been shown to be capable of analyzing unstructured data and correctly estimating the likelihood of patients getting different illnesses, including cancer. […] AI models can predict the risk of illness recurrence following a therapeutic option. […] The applications of AI for the prediction of cancer recurrence have demonstrated greater accuracy than standard statistical models, which will further facilitate the optimization of clinical follow-up plans.
#11 Application of Deep Learning in Cancer Prognosis Prediction Model
https://pmc.ncbi.nlm.nih.gov/articles/PMC10503281/
As an important branch of artificial intelligence and machine learning, deep learning (DL) has been widely used in various aspects of cancer auxiliary diagnosis, among which cancer prognosis is the most important part. High-accuracy cancer prognosis is beneficial to the clinical management of patients with cancer. Compared with other methods, DL models can significantly improve the accuracy of prediction. Therefore, this article is a systematic review of the latest research on DL in cancer prognosis prediction. […] Accurate cancer prognosis could help improve the cure rate and overall survival of patients with cancer. […] At present, cancer prognosis prediction methods mainly include traditional machine learning, radiomics, and DL. […] DL is a relatively novel method that uses structures superimposed by multilayer neural networks to train and predict cancer-related data sets.
#12 Application of Deep Learning in Cancer Prognosis Prediction Model
https://pmc.ncbi.nlm.nih.gov/articles/PMC10503281/
The above study shows that DNNs can better integrate multiomics data, including but not limited to clinical, pathological, and radiomics data, and have higher accuracy than traditional machine learning models in predicting cancer survival. […] The results suggest that CNN image-based survival prediction is promising and could be integrated with existing clinical data. […] The combination of unsupervised and supervised learning gives DBNs a great deal of flexibility. […] DRNs have been a hot research topic in the field of cancer prognosis prediction, and it has been widely used in 3D medical images. […] The development of ViT provides a new research scheme for the prognosis prediction of cancer. […] The above studies demonstrated the powerful performance of building a ViT cancer survival prognostic model. […] The research of the DL cancer prognostic model could better cure patients with cancer.
#13 Application of Deep Learning in Cancer Prognosis Prediction Model
https://pmc.ncbi.nlm.nih.gov/articles/PMC10503281/
The above study shows that DNNs can better integrate multiomics data, including but not limited to clinical, pathological, and radiomics data, and have higher accuracy than traditional machine learning models in predicting cancer survival. […] The results suggest that CNN image-based survival prediction is promising and could be integrated with existing clinical data. […] The combination of unsupervised and supervised learning gives DBNs a great deal of flexibility. […] DRNs have been a hot research topic in the field of cancer prognosis prediction, and it has been widely used in 3D medical images. […] The development of ViT provides a new research scheme for the prognosis prediction of cancer. […] The above studies demonstrated the powerful performance of building a ViT cancer survival prognostic model. […] The research of the DL cancer prognostic model could better cure patients with cancer.
#14 Application of Deep Learning in Cancer Prognosis Prediction Model
https://pmc.ncbi.nlm.nih.gov/articles/PMC10503281/
The above study shows that DNNs can better integrate multiomics data, including but not limited to clinical, pathological, and radiomics data, and have higher accuracy than traditional machine learning models in predicting cancer survival. […] The results suggest that CNN image-based survival prediction is promising and could be integrated with existing clinical data. […] The combination of unsupervised and supervised learning gives DBNs a great deal of flexibility. […] DRNs have been a hot research topic in the field of cancer prognosis prediction, and it has been widely used in 3D medical images. […] The development of ViT provides a new research scheme for the prognosis prediction of cancer. […] The above studies demonstrated the powerful performance of building a ViT cancer survival prognostic model. […] The research of the DL cancer prognostic model could better cure patients with cancer.
#15
https://link.springer.com/article/10.1007/s13167-010-0044-z
The identification of pattern-based biomarkers for prognosis is a major field of current clinical research in many cancer types including breast cancer. […] Consequently, the identification of prognostic markers identifying the subset of patients eligible for a watchful waiting procedure and/or adjuvant anti-hormonal/anti-HER2 therapy could help to minimize therapy-induced side effects. […] The first breast cancer prognostic signature to be described has been a 70-gene signature by van t Veer et al. […] The 70-gene prognosis profile was a strong predictor of the development of distant metastasis in patients with both lymph-node-negative as well as lymph-node-positive disease. […] The authors then compared the probability that patients classified according to either the 70-gene expression profile, to St. Gallen criteria, or to the National Institutes of Health (NIH) consensus criteria would remain free of distant metastasis.
#16
https://link.springer.com/article/10.1007/s13167-010-0044-z
The 70-gene signature classified 41% of patients among lymph-node positive patients into a low-risk group for metastasis. […] The 70-gene profile illustrates a high power in respect to its negative predicting value, further validated in an independent study of 100 older lymph-node negative patients. […] The 70-gene signature not only correlates with established factors of prognosis like age, grading and ER-status, but outperforms the well-established prognostic algorithms such as the St. Gallen criteria. […] The 70-gene-profile reliably identifies high-risk patients that require chemotherapy. […] Even more important, the 70-gene-profile identifies patients with low risk of recurrence, who could spare chemotherapy. […] The 21-gene recurrence score adds important information to the traditional pathological approaches, but some problems remain.
#17
https://link.springer.com/article/10.1007/s13167-010-0044-z
The 70-gene signature classified 41% of patients among lymph-node positive patients into a low-risk group for metastasis. […] The 70-gene profile illustrates a high power in respect to its negative predicting value, further validated in an independent study of 100 older lymph-node negative patients. […] The 70-gene signature not only correlates with established factors of prognosis like age, grading and ER-status, but outperforms the well-established prognostic algorithms such as the St. Gallen criteria. […] The 70-gene-profile reliably identifies high-risk patients that require chemotherapy. […] Even more important, the 70-gene-profile identifies patients with low risk of recurrence, who could spare chemotherapy. […] The 21-gene recurrence score adds important information to the traditional pathological approaches, but some problems remain.
#18
https://link.springer.com/article/10.1007/s13167-010-0044-z
The Oncotype DX assay is, although prognostic and predictive before 5 years, not prognostic and predictive after 5 years. […] The HOXB13:IL17BR expression ratio is a prognostic factor in early-stage breast cancer. […] The predictive power of the HDPP signature in trastuzumab treatment was only investigated in a small set of 22 patients treated neoadjuvantly with trastuzumab and vinorelbine. […] The expression of ER for example is negatively predictive for response to chemotherapy and the same might be suggested for its accompanying luminal A gene expression profile. […] The application of tumor intrinsic subtypes in predicting response to chemotherapy might be used in settings, where it is still difficult to find gene expression patterns that predict for therapy response. […] In conclusion, Patients that harbor a prognostic high-risk gene expression profile according to Fig. 3 and are assigned to undergo chemotherapy to minimize recurrence rate might be further allocated to tumor-specific chemotherapy.
#19 A narrative review of prognosis prediction models for non-small cell lung cancer: what kind of predictors should be selected and how to improve models?
https://atm.amegroups.org/article/view/81325/html
Another significant prognostic biomarker resulting from genomics analysis is tumor mutational burden (TMB), which is defined as the number of nonsynonymous mutations, especially in immunotherapy patients. […] The wide diversity of tumor mutations is a great challenge for researchers developing effective biomarkers for diagnosis or prognosis because of the large proportions of the genome needed to be examined to provide adequate sensitivity. […] Increasingly, evidence has shown that molecular biomarkers can greatly benefit prognostic prediction. […] The most fundamental role of the clinical prognosis prediction model is to allow clinicians and patients to understand the prognosis of patients simply, quickly, and effectively, and to guide treatment in a timely manner. […] The wide application of next-generation sequencing technology provides a variety of predictive factors for clinical prediction models. However, it seems that gene-related biomarkers cannot obviously improve the performance of models. An urgent solution as to how to simplify the required biomarkers while ensuring stability is required.
#20 A narrative review of prognosis prediction models for non-small cell lung cancer: what kind of predictors should be selected and how to improve models?
https://atm.amegroups.org/article/view/81325/html
Another significant prognostic biomarker resulting from genomics analysis is tumor mutational burden (TMB), which is defined as the number of nonsynonymous mutations, especially in immunotherapy patients. […] The wide diversity of tumor mutations is a great challenge for researchers developing effective biomarkers for diagnosis or prognosis because of the large proportions of the genome needed to be examined to provide adequate sensitivity. […] Increasingly, evidence has shown that molecular biomarkers can greatly benefit prognostic prediction. […] The most fundamental role of the clinical prognosis prediction model is to allow clinicians and patients to understand the prognosis of patients simply, quickly, and effectively, and to guide treatment in a timely manner. […] The wide application of next-generation sequencing technology provides a variety of predictive factors for clinical prediction models. However, it seems that gene-related biomarkers cannot obviously improve the performance of models. An urgent solution as to how to simplify the required biomarkers while ensuring stability is required.
#21 Automated real-world data integration improves cancer outcome prediction | Nature
https://www.nature.com/articles/s41586-024-08167-5
Models incorporating more detailed clinical, genomic, radiomic and histopathologic data have shown promise in better risk stratification, although these efforts frequently rely on and are limited by manual extraction of key data elements and are studied in cohorts of modest size. […] Thus, MSK-CHORD’s size enables adequately powered identification of post-treatment mutations across multiple cancers, and NLP-derived prior treatment is an important complement to institutional treatment records in such analyses. […] Our results indicate that multimodal biomarkers are superior to disease stage for prognostication. […] Thus, our cohort allows for validation of proposed genomic-histopathologic associations. […] Overall, our analysis confirms several genomic-metastasis site associations observed in smaller or non-temporal cohorts but also identifies new potential genomic changes of prognostic importance that can be prospectively validated.
#22 Automated real-world data integration improves cancer outcome prediction | Nature
https://www.nature.com/articles/s41586-024-08167-5
The digitization of health records and growing availability of tumour DNA sequencing provide an opportunity to study the determinants of cancer outcomes with unprecedented richness. […] Leveraging MSK-CHORD to train machine learning models to predict overall survival, we find that models including features derived from natural language processing, such as sites of disease, outperform those based on genomic data or stage alone as tested by cross-validation and an external, multi-institution dataset. […] We demonstrate the feasibility of automated annotation from unstructured notes and its utility in predicting patient outcomes. […] The resulting data are provided as a public resource for real-world oncologic research. […] In this study, we used a large, integrated dataset to develop improved models of cancer outcome.
#23 Automated real-world data integration improves cancer outcome prediction | Nature
https://www.nature.com/articles/s41586-024-08167-5
Models incorporating more detailed clinical, genomic, radiomic and histopathologic data have shown promise in better risk stratification, although these efforts frequently rely on and are limited by manual extraction of key data elements and are studied in cohorts of modest size. […] Thus, MSK-CHORD’s size enables adequately powered identification of post-treatment mutations across multiple cancers, and NLP-derived prior treatment is an important complement to institutional treatment records in such analyses. […] Our results indicate that multimodal biomarkers are superior to disease stage for prognostication. […] Thus, our cohort allows for validation of proposed genomic-histopathologic associations. […] Overall, our analysis confirms several genomic-metastasis site associations observed in smaller or non-temporal cohorts but also identifies new potential genomic changes of prognostic importance that can be prospectively validated.
#24 Predicting cancer prognosis and drug response from the tumor microbiome | Nature Communications
https://www.nature.com/articles/s41467-022-30512-3
Tumor gene expression is predictive of patient prognosis in some cancers. […] Here, we show that tumor microbial abundances, alone or in combination with tumor gene expression, can predict cancer prognosis and drug response to some extent microbial abundances are significantly less predictive of prognosis than gene expression, although similarly as predictive of drug response, but in mostly different cancer-drug combinations. […] We show that in four cancer types, adrenocortical carcinoma, cervical squamous cell carcinoma, brain lower grade glioma, and subcutaneous skin melanoma, tumor microbial abundances are better predictors of prognosis than clinical covariates alone. However, we find that tumor gene expression is a more powerful predictor of prognosis, across a wider range of cancer types, than microbial abundances.
#25 Predicting cancer prognosis and drug response from the tumor microbiome | Nature Communications
https://www.nature.com/articles/s41467-022-30512-3
Tumor gene expression is predictive of patient prognosis in some cancers. […] Here, we show that tumor microbial abundances, alone or in combination with tumor gene expression, can predict cancer prognosis and drug response to some extent microbial abundances are significantly less predictive of prognosis than gene expression, although similarly as predictive of drug response, but in mostly different cancer-drug combinations. […] We show that in four cancer types, adrenocortical carcinoma, cervical squamous cell carcinoma, brain lower grade glioma, and subcutaneous skin melanoma, tumor microbial abundances are better predictors of prognosis than clinical covariates alone. However, we find that tumor gene expression is a more powerful predictor of prognosis, across a wider range of cancer types, than microbial abundances.
#26 Outcome Prediction Using Multi-Modal Information: Integrating Large Language Model-Extracted Clinical Information and Image Analysis
https://www.mdpi.com/2072-6694/16/13/2402
Outcome Prediction Using Multi-Modal Information: Integrating Large Language Model-Extracted Clinical Information and Image Analysis […] Predicting the survival of bladder cancer patients following cystectomy can offer valuable information for treatment planning, decision-making, patient counseling, and resource allocation. […] This study demonstrates the potential of employing LLMs to process medical data, and of integrating LLM-processed data into modeling for prognosis. […] Survival prediction post-cystectomy is essential for the follow-up care of bladder cancer patients. […] The addition of imaging information can further improve the survival prediction accuracy of the models. […] Numerous studies have shown that survival of bladder cancer patients may be predicted using clinicopathological information, histology, genetics, and molecular markers. […] The rapid development of LLMs provides an opportunity to explore the use of LLMs for the automated extraction of information from clinical reports, which can be used as input to both nomogram models and integrated models, combining clinical and image information for survival prediction.
#27 Outcome Prediction Using Multi-Modal Information: Integrating Large Language Model-Extracted Clinical Information and Image Analysis
https://www.mdpi.com/2072-6694/16/13/2402
This study shows that LLMs can extract clinical information accurately from a patientâs medical records and a patientâs medical reports. […] The study also demonstrates the efficacy of combining automatically extracted clinical and imaging features to improve the performance of survival predictive models for bladder cancer patients.
#28 Predict Breast
https://breast.predict.cam/
Predict is an online tool that helps patients and clinicians see how different treatments for early invasive breast cancer might improve survival rates after surgery. […] It then uses data about the survival of similar women in the past to show the likely proportion of such women expected to survive up to fifteen years after their surgery with different treatment combinations. […] Patients should use it in consultation with a medical professional.
#29
https://link.springer.com/article/10.1007/s10549-025-07654-1
Outcome prediction research in early-onset breast cancer (EoBC) is limited. […] This study evaluated the predictive performance of NHS PREDICT v2.1 and developed two prediction models for 5-year and 10-year all-cause mortality in a cohort of EoBC patients in Alberta, Canada. […] PREDICT v2.1 tended to overestimate 5-year mortality in those with 30% predicted risks and 10-year mortality in those with 50% predicted risks for EoBC in Alberta, Canada. […] The current study assessed the predictive performance of PREDICT’s 5-year and 10-year all-cause mortality estimates in a real-world population of patients with EoBC in Alberta, Canada. […] Our study reports several calibration measures for PREDICT in the EoBC setting. […] Overestimation of 5-year mortality was observed in ER-positive, HER2-positive, grade III, and T3 disease.
#30 Machine learning-based prediction of survival prognosis in cervical cancer | BMC Bioinformatics | Full Text
https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-021-04261-x
Accurately forecasting the prognosis could improve cervical cancer management, however, the currently used clinical features are difficult to provide enough information. […] A miRNAs-based machine learning cervical cancer survival prediction model was developed that robustly stratifies cervical cancer patients into high survival rate (5-years survival rate90%), moderate survival rate (5-years survival rate 65%), and low survival rate (5-years survival rate40%). […] Survival prediction after first diagnosis is important for both disease specialist and patients or their family members. […] Accurately forecasting the survival cancer patients are important for therapeutic decision. Currently, most molecular-based survival prediction model stratified the patients into two groups with different survival outcome.
#31 Machine learning-based prediction of survival prognosis in cervical cancer | BMC Bioinformatics | Full Text
https://bmcbioinformatics.biomedcentral.com/articles/10.1186/s12859-021-04261-x
In this study, the 10 miRNAs-based prediction model developed by SVM program could robustly stratify the cervical patients into three groups (5-years survival rate90%, 65% and40%), which significantly improves the usefulness of the model. […] Collectively, a miRNAs-based ML CCSPM that stratifies cervical cancer patients into high survival rate (5-years survival rate90%), moderate survival rate (5-years survival rate 65%) and low survival rate (5-years survival rate40%) was developed.
#32 ATM mutations alone do not predict early treatment need in chronic lymphocytic leukemia, study finds
https://medicalxpress.com/news/2025-05-atm-mutations-early-treatment-chronic.html
In chronic lymphocytic leukemia (CLL), certain recurrent genetic alterations are known to influence disease progression and survival. […] In a study published in Leukemia involving 3,631 untreated CLL patientsâcollected from several European centers and coordinated by the European Research Initiative on CLL (ERIC) as part of the HARMONY Allianceâresearchers from Karolinska Institutet explored the prognostic significance of ATM mutations, along with mutations in nine additional genes. Prognosis was measured based on time to first treatment (TTFT). […] Furthermore, patients with any ATM-related abnormalities (ATM mutations and/or del(11q)) experienced a significantly shorter TTFT. However, after adjusting for other genetic factors, only del(11q)âand not ATM mutations aloneâemerged as an independent predictor of earlier need for treatment.
#33 ATM mutations alone do not predict early treatment need in chronic lymphocytic leukemia, study finds
https://medicalxpress.com/news/2025-05-atm-mutations-early-treatment-chronic.html
„In short, our findings strengthen the role of del(11q)âand not ATM mutations aloneâas a critical marker for higher risk of progression in patients with CLL. This insight can help doctors to improve risk stratification and personalize treatment decisions for patients,” says Birna Thorvaldsdottir, post-doc at the Department of Molecular Medicine and Surgery.
#34 [2306.14596] Deep Learning for Cancer Prognosis Prediction Using Portrait Photos by StyleGAN Embedding
https://arxiv.org/abs/2306.14596
Survival prediction for cancer patients is critical for optimal treatment selection and patient management. […] In this work, the efficacy of objectively capturing and using prognostic information contained in conventional portrait photographs using deep learning for survival predication purposes is investigated for the first time. […] Utilizing the state-of-the-art survival analysis models and based on StyleGAN’s latent space photo embeddings, this approach achieved a C-index of 0.677, which is notably higher than chance and evidencing the prognostic value embedded in simple 2D facial images. […] Moreover, a health attribute is obtained from regression coefficients, which has important potential value for patient care.
#35 Google Goes Cancer: Improving Outcome Prediction for Cancer Patients by Network-Based Ranking of Marker Genes | PLOS Computational Biology
https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1002511
Predicting the clinical outcome of cancer patients based on the expression of marker genes in their tumors has received increasing interest in the past decade. […] Accurate predictors of outcome and response to therapy could be used to personalize and thereby improve therapy. […] Our approach ranks genes according to their prognostic relevance using both expression and network information in a manner similar to Google’s PageRank. […] We applied this method to gene expression profiles which we obtained from 30 patients with pancreatic cancer, and identified seven candidate marker genes prognostic for outcome. […] Notably, signatures derived from our candidate markers were independently predictive of outcome and superior to established clinical prognostic factors such as grade, tumor size, and nodal status.
#36 Google Goes Cancer: Improving Outcome Prediction for Cancer Patients by Network-Based Ranking of Marker Genes | PLOS Computational Biology
https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1002511
The aim of our study was therefore (i) to carry out a genome-wide screen for genes whose expression in pancreatic cancer tissue samples reliably correlates with the patient survival time, and (ii) to use these genes as a molecular signature for reliable survival prediction. […] We found that signatures based on these markers were more accurate than traditional clinical parameters and more accurate than signatures identified with other computational approaches. […] Both signatures improve prediction of patient prognosis compared to the use of clinical parameters when used for immunohistochemical staining of the tumor tissue. […] The additional predictive value of the signature markers compared to clinical parameters was 9% for patients with and 6% for patients without adjuvant therapy. […] Hence, the markers found in our study are independently predictive of outcome and superior to established clinical parameters.
#37 Prognostic models for outcome prediction in patients with advanced hepatocellular carcinoma treated by systemic therapy: a systematic review and critical appraisal | BMC Cancer | Full Text
https://bmccancer.biomedcentral.com/articles/10.1186/s12885-022-09841-5
The most common indicators for evaluating the predictive performance of a prognostic model were discrimination and calibration. […] Thirty-two prognostic models were externally validated. […] All models had a high risk of bias, which may limit their application in clinical practice. […] The development and validation of these models will aid the identification of patients with HCC who may benefit from systemic therapy, and guide treatment. […] Future studies should focus on updating existing prognosis models by adjusting predictors to improve performance and promoting their clinical practice through external validation.
#38 Prognostic models for outcome prediction in patients with advanced hepatocellular carcinoma treated by systemic therapy: a systematic review and critical appraisal | BMC Cancer | Full Text
https://bmccancer.biomedcentral.com/articles/10.1186/s12885-022-09841-5
The most common indicators for evaluating the predictive performance of a prognostic model were discrimination and calibration. […] Thirty-two prognostic models were externally validated. […] All models had a high risk of bias, which may limit their application in clinical practice. […] The development and validation of these models will aid the identification of patients with HCC who may benefit from systemic therapy, and guide treatment. […] Future studies should focus on updating existing prognosis models by adjusting predictors to improve performance and promoting their clinical practice through external validation.
#39 A narrative review of prognosis prediction models for non-small cell lung cancer: what kind of predictors should be selected and how to improve models?
https://atm.amegroups.org/article/view/81325/html
Another significant prognostic biomarker resulting from genomics analysis is tumor mutational burden (TMB), which is defined as the number of nonsynonymous mutations, especially in immunotherapy patients. […] The wide diversity of tumor mutations is a great challenge for researchers developing effective biomarkers for diagnosis or prognosis because of the large proportions of the genome needed to be examined to provide adequate sensitivity. […] Increasingly, evidence has shown that molecular biomarkers can greatly benefit prognostic prediction. […] The most fundamental role of the clinical prognosis prediction model is to allow clinicians and patients to understand the prognosis of patients simply, quickly, and effectively, and to guide treatment in a timely manner. […] The wide application of next-generation sequencing technology provides a variety of predictive factors for clinical prediction models. However, it seems that gene-related biomarkers cannot obviously improve the performance of models. An urgent solution as to how to simplify the required biomarkers while ensuring stability is required.
#40 A narrative review of prognosis prediction models for non-small cell lung cancer: what kind of predictors should be selected and how to improve models?
https://atm.amegroups.org/article/view/81325/html
Another significant prognostic biomarker resulting from genomics analysis is tumor mutational burden (TMB), which is defined as the number of nonsynonymous mutations, especially in immunotherapy patients. […] The wide diversity of tumor mutations is a great challenge for researchers developing effective biomarkers for diagnosis or prognosis because of the large proportions of the genome needed to be examined to provide adequate sensitivity. […] Increasingly, evidence has shown that molecular biomarkers can greatly benefit prognostic prediction. […] The most fundamental role of the clinical prognosis prediction model is to allow clinicians and patients to understand the prognosis of patients simply, quickly, and effectively, and to guide treatment in a timely manner. […] The wide application of next-generation sequencing technology provides a variety of predictive factors for clinical prediction models. However, it seems that gene-related biomarkers cannot obviously improve the performance of models. An urgent solution as to how to simplify the required biomarkers while ensuring stability is required.
#41 A narrative review of prognosis prediction models for non-small cell lung cancer: what kind of predictors should be selected and how to improve models?
https://atm.amegroups.org/article/view/81325/html
Another significant prognostic biomarker resulting from genomics analysis is tumor mutational burden (TMB), which is defined as the number of nonsynonymous mutations, especially in immunotherapy patients. […] The wide diversity of tumor mutations is a great challenge for researchers developing effective biomarkers for diagnosis or prognosis because of the large proportions of the genome needed to be examined to provide adequate sensitivity. […] Increasingly, evidence has shown that molecular biomarkers can greatly benefit prognostic prediction. […] The most fundamental role of the clinical prognosis prediction model is to allow clinicians and patients to understand the prognosis of patients simply, quickly, and effectively, and to guide treatment in a timely manner. […] The wide application of next-generation sequencing technology provides a variety of predictive factors for clinical prediction models. However, it seems that gene-related biomarkers cannot obviously improve the performance of models. An urgent solution as to how to simplify the required biomarkers while ensuring stability is required.