Following and predicting chemotherapy side effects on mobile devices with machine learning

Bolnikova ocena zdravstvenega stanja postaja vedno bolj pomembna. Pri oceni kakovosti življenja bolniki z rakom sami, brez posredovanja zdravnika ali druge osebe, podajo podatke o simptomih bolezni in stranskih učinkih zdravljenja. Z namenom njihovega sprotnega beleženja in boljšega obvladovanja smo leta 2019 razvili mobilno aplikacijo mPRO Mamma, ki se je s tem ciljem uporabljala na Onkološkem inštitutu Ljubljana v okviru prospektivne raziskave pri bolnicah z rakom dojk. Raziskava je pokazala izboljšanje nekaterih vidikov kakovosti življenja v primerjavi s kontrolno skupino, ki mobilne aplikacije ni uporabljala. Na podlagi omejitev takratne različice aplikacije na uporabnost samo za eno vrsto raka, smo se odločili za njeno nadgradnjo. Tako je nastala mobilna aplikacija OnkoVed, ki podpira večje število rakavih obolenj, dodatno smo digitalizirali proces spremljanja zdravljenja, vključili nekatere uveljavljene vprašalnike in zgradili napovedne modele, ki bi na podlagi vnosov skušali prepoznati dneve poslabšanja stanja. Pri našem pristopu smo uporabili nevronske mreže tipa LSTM in odločitvena drevesa tipa XGBoost. Z zgrajenimi modeli pa smo zgolj na podlagi dnevnih vnosov jakosti stranskih učinkov našli le šibko povezavo s prihajajočim poslabšanjem stanja.Patient reported outcomes are becoming increasingly more important. In order for patients to assess their quality of life, they themselves (without the medical oversee) mark their experienced side effects as a result of their treatment. In 2019 we developed a mobile application called mPRO Mamma with the goal of supporting this process digitally, which has been used in a prospective study on Institute of Oncology Ljubljana, which targeted breast cancer patients. Study showed an increase in quality of life of patients that were using the application in contrast with the control group. Based on the limitations of the before mentioned application to just one type of cancer we identified required improvements. We developed a mobile application called OnkoVed, which now supports various cancer types. We also additionally digitized the treatment process via the inclusion of standardised questionnaires and developed machine learning models to be able to predict days of patient health deterioration. In our approach we used neural networks of type LSTM and XGBoost decision trees. With the built models we were only able to find a weak link between the obtained daily data of perceived side effects and the days of health deterioration