Logo KIN

AI-basert beslutningsstøtte for ryggkirurgi


AI-basert beslutningsstøtte for ryggkirurgi basert på data fra Nasjonalt kvalitetsregister for ryggkirurgi og strukturerte data fra elektronisk pasientjournal

Consortium for Patient Centered Artificial Intelligence:

Project aim and objectives

The overall aim of this large-scale interdisciplinary research consortium is to catalyze a fundamental paradigm shift to change daily clinical practice by developing and implementing AI as an integral part of real-time diagnosis, treatment, and follow-up of patients in a university hospital. To achieve the overall aim, the project’s objectives are to:

  • Develop and implement safe and effective real-time clinical decision support (CDS) tools for outcomes prediction, including postoperative, therapeutic acute and late adverse events using the information embedded in clinical quality registries and by live sampling, real-time, from the EHRs.

Research hypothesis

We hypothesize that

  • CDS- and early warning-systems based on data from EHRs and quality registers, leveraging new AI methodology, has the potential to change clinical practice and be implemented for diagnosis, treatment, and quality improvements.
  • Multidisciplinary AI based research can be implemented into clinical use and break the barriers to deployment, by sufficiently fulfilling the organizational conditions in the implementing process as well as the organization`s degree of readiness for change.

WP 3 AI-derived clinical decision support base don registry data

National clinical quality registers contain comprehensive data about most patients being treated for a specific condition or undergoing specific interventions. However, this information is underutilized in shared decision making for individual patients. The Norwegian registry for spine surgery (NORspine) annually registers more than 6 000 patients undergoing surgical treatment for neck- or back-pain radiating to the arm or leg. Approximately 25 % do not improve, and a small proportion experience worsening.

Goals

  1. Develop real-time CDS tool for selection of patients to surgery, integrate it into the EHR and implement it in routine shared decision making at an outpatient clinic.
  2. Perform prospective trials to study effectiveness and safety of our CDS tool in compliance with guidelines for AI clinical research.

Data

NORspine contains detailed pre-operative physician and patient reported data and patient reported outcome measures (PROM’s) from more than 60 000 patients. Approval has been granted by REK and all data are already available. The registry is continuously updated with more than 6 000 new entries annually.

Additionally, we will exploit structured EHR data such as demographic data, medication and codes.

Approach

The team has established and validated cut-off criteria to identify patients experiencing failure of treatment or worsening. We have also developed a predictive model, based on pre-operative data, using classical biostatistics (multivariate analysis), to calculate individual pre-treatment risk for these unfavorable outcomes. The performance of the model is promising, but not satisfactory for real-time clinical use.

Therefore, the first step in this WP is to develop an improved ML-derived predictive model based on all the available data, which can form the basis of a real-time CDS tool for spine surgery. Many of the challenges related to the ML methodology are shared with other work-packages. Hence, there will be synergies. To ensure both safety and effectiveness, we will focus on interpretability and we will develop ML methodology for heterogeneous data sources which is capable of exploiting contextual and prior information.

Team

Clinical ph.d.-student (to be recruited)

Tore Solberg, professor (NORSpine, neurosurgery)

Tor Ingebrigtsen, professor (neurosurgery)

Collaboration partners

National Center for eHealth Research (NSe)

Case-study

WP3 will serve as a case for the research in WP5 Comparative and action-based research on implementation of AI into clinical practice.

Samarbeidspartnere

UiT, E-helseforskning

Formål

Å utvikle beslutningsstøtte for utvelgelse av pasienter som vil ha nytte av ryggkirurgi.

Effekt/kvalitetsmål

Pasientrapportert utfall etter ryggkirurgi.

Problemstilling

Om lag 25% av pasienter som opereres i ryggen har utilfredsstillende utfall (liten/ingen bedring eller forverring). Vi forventer at AI-basert beslutningsstøtte vil forbedre utvelgelsen av pasienter til operasjon og dermed utfallet for dem som blir operert.


Pågående prosjekt

Prosjektperiode

2022 - 2028

Kategorier

Fokusområde:

Klinisk, IKT-infrastruktur/datatilgang

Type helsetjeneste:

Spesialist

Type data:

Strukturerte data

Datakilde:

Selvrapportert, Journal, Register

Planlagt sluttfase:

Implementering

Oppgave:

Behandlingsvalg

Pasientgruppe

Pasienter som er henvist til vurdering før eventuell ryggkirurgi.

Pasientvolum/datamengde

Vi skal bruke hele erfaringsgrunnlaget i Nasjonalt kvalitetsregister for ryggkirurgi (ca. 70'000 pasienter) som utgangspunkt, og utvikle en algoritme som oppdateres kontinuerlig etter hvert som nye pasienter registreres (ca. 5000 pr. år). Uttesting av beslutningsstøtten vil først skje i en liten pilotstudie på UNN og deretter breddes ut til flere norske sykehus og til slutt eventuelt bli testet i en RCT.

Prosjekteier

UNN

Helseregion

Helse Nord

Kontaktpersoner