Artificial intelligence and electronic health records could help suicide prevention efforts

Each year around 800,000 people across the world take their own lives, leading to a huge impact on family, the community and health professionals.

 

Unfortunately it is not easy to identify those at risk of suicide. A large number of variables and complex interactions mean only a trained clinician can assess for immediate risk of suicide.

 

However, the majority of people who take their own lives attend services other than mental health in their final year for apparently unrelated reasons. Now new research has looked at ways of using this health information to identify the most vulnerable.

 

The research was carried out by a team including Dr Marcos del Pozo Banos, Professor Ann John, Professor Damon Berridge and Professor Keith Lloyd from Swansea University’s Medical School, and Dr Caroline Jones of the University’s Hillary Rodham Clinton School of Law, in collaboration with academic colleagues from European universities as well as Dr Sarah Spencer, emergency medicine consultant at Princess of Wales Hospital in Bridgend.

 

Their findings, which have just been published, revealed that more than 80 per cent of suicide cases studied had had at least one contact with their GP during their last year.

 

Dr del Pozo Banos said: “We wanted to see if we could develop an algorithm that analyses routinely collected health data to flag people, so that when patients present with seemingly unrelated conditions, practitioners can ask them appropriate questions about their thoughts and feelings if required. ”

 

He explained the crucial role artificial intelligence (AI) can now play in processing millions of records with thousands of variables to build a risk model very quickly. It can then process the health history of patients and highlight those who may be at risk.

 

Dr del Pozo Banos, an expert in mental health informatics and AI, said that by using primary care data it was possible to substantially increase the coverage of the system compared to other published proposals.

The research was co-authored by Professor Nicolai Petkov, from the University of Groningen, Netherlands; Kate Southern, from Cardiff Adult Self Injury Project; and Professor Carlos Manuel Travieso, of the University of Las Palmas de Gran Canaria, Spain.

 

It used the University’s SAIL Databank to obtain information about more than 2,000 suicides held in SID-Cymru – an electronic cohort of all deaths by suicide and of undetermined intent in Wales since 2001 – plus 20 controls per case.

 

Dr del Pozo Banos described their preliminary results, which recorded almost 75 per cent accuracy, as very promising.

 

The team now plans to gradually increase the complexity of the AI system to improve the tool’s precision before carrying out thorough trials.

 

He said: “Our proposal will not replace clinical assessment of immediate risk, but we hope it will help clinicians to identify vulnerable people when they access health services so that the right questions may be asked.

 

“We are a long way from employing such methods in clinical practice, but this is a first step to harnessing the potential of routinely collected electronic health records to support clinical practice in real time.”

 

Professor John added: “AI methods have huge potential to improve suicide prevention, and this is an exciting development.

 

“However, we are still at the early development stages before any practical use. Collaborations between clinicians, academics (both from a traditional epidemiological background and data scientists), ethical experts, patients and people who use services are vital for the development of these innovative tools.”

 

Gallai deallusrwydd artiffisial a chofnodion iechyd electronig helpu gydag ymdrechion i atal hunanladdiad

Bob blwyddyn mae tuag 800,000 o bobl ledled y byd yn eu lladd eu hunain, sy’n cael effaith enfawr ar deuluoedd, y gymuned ac ymarferwyr iechyd proffesiynol. 

 

Yn anffodus nid yw’n hawdd adnabod y rhai mewn perygl o hunanladdiad. O ganlyniad i nifer fawr o newidynnau a rhyngweithiadau cymhleth, clinigwyr hyfforddedig yn unig all asesu perygl dybryd o hunanladdiad.

 

Fodd bynnag, mae’r rhan fwyaf o bobl sy’n eu lladd eu hunain yn mynychu gwasanaethau heblaw am rai iechyd meddwl yn eu blwyddyn derfynol am resymau nad ydynt i bob golwg yn gysylltiedig. Bellach mae ymchwil newydd wedi edrych ar ffyrdd o ddefnyddio’r wybodaeth iechyd hon i adnabod y bobl sy’n wynebu’r perygl mwyaf.

 

Cynhaliwyd yr ymchwil gan dîm yn cynnwys Dr Marcos del PozoBanos, yr Athro Ann John, yr Athro Damon Berridge a’r Athro Keith Lloyd o Ysgol Feddygol Prifysgol Abertawe , a Dr Caroline Jones o Ysgol y Gyfraith HillaryRodham Clinton, ar y cyd â chydweithwyr academaidd o brifysgolion yn Ewrop yn ogystal â Dr Sarah Spencer, ymgynghorydd meddygaeth frys yn Ysbyty Tywysoges Cymru ym Mhen-y-Bont.

 

Datgelodd eu canfyddiadau, sydd newydd gael eu cyhoeddi, fod dros 80 y cant o’r achosion hunanladdiad a astudiwyd wedi cael cyswllt o leiaf unwaith gyda’u meddyg teulu yn ystod eu blwyddyn derfynol.

 

Meddai Dr del PozoBanos: “Roeddem am weld a allem ddatblygu algorithm sy’n dadansoddi data iechyd a gesglir yn rheolaidd i dynnu sylw at bobl, er mwyn i ymarferwyr ofyn cwestiynau priodol ynghylch meddyliau a theimladau i gleifion sy’n dod i’w gweld gyda chyflyrau nad ydynt, i bob golwg, yn gysylltiedig.”

 

Esboniodd y rôl hollbwysig y gall deallusrwydd artiffisial ei chwarae bellach wrth brosesu miliynau o gofnodion gyda miloedd o amrywiannau i greu model risgiau’n gyflym iawn. Yna gall brosesu hanes iechyd cleifion a thynnu sylw at y rhai a allai fod mewn perygl.

 

Dywedodd Dr del PozoBanos, sy’n arbenigwr mewn gwybodeg iechyd meddwl a deallusrwydd artiffisial, fod modd defnyddio data gofal sylfaenol i gynyddu cwmpas y system yn sylweddol o’i chymharu â chynigion eraill a gyhoeddwyd drwy ddefnyddio data gofal sylfaenol.

 

Cafodd yr ymchwil ei hysgrifennu ar y cyd gan yr Athro NicolaiPetkov, o BrifysgolGroningen, yr Iseldiroedd; Kate Southern, o Brosiect Hunan-niwed Oedolion Caerdydd; a’r Athro  CarlosManuelTravieso, o Brifysgol Las Palmas de GranCanaria, Sbaen.

 

Defnyddiodd Banc Data SAIL y Brifysgol i gael gwybodaeth am dros 2,000 o hunanladdiadau a ddelir yn SID-Cymru – carfan electronig o’r holl farwolaethau drwy hunanladdiad a bwriad amhendant yng Nghymru ers 2001 – yn ogystal ag 20 o reolyddion fesul achos.

 

 

Disgrifiodd Dr del PozoBanos eu canlyniadau rhagarweiniol, a gofnodwyd cywirdeb o bron 75 y cant, fel canlyniadau addawol iawn.

 

Mae’r tîm bellach yn cynllunio cynyddu cymhlethdod y system deallusrwydd artiffisial yn raddol er mwyn gwella manwl gywirdeb yr offeryn cyn cynnal treialon trylwyr.

 

Meddai: “Ni fydd ein cynnig yn disodli’r gwaith o asesu risg dybryd, ond gobeithiwn y bydd yn helpu clinigwyr i adnabod pobl sydd mewn perygl o hunanladdiad pan fyddant yn defnyddio gwasanaethau iechyd i sicrhau y gofynnir y cwestiynau cywir.

 

“Rydym yn bell i ffwrdd o ddefnyddio dulliau o’r fath mewn ymarfer clinigol, ond dyma’r cam cyntaf tuag at harneisio potensial cofnodion iechyd electronig a gesglir yn rheolaidd er mwyn cefnogi ymarfer clinigol mewn amser real.”

 

Ychwanegodd yr Athro John: “Mae gan ddulliau deallusrwydd artiffisial botensial enfawr i wella atal hunanladdiad, ac mae hwn yn ddatblygiad cyffrous.

 

“Fodd bynnag, rydym yn dal i fod yn y camau datblygu cynnar cyn unrhyw ddefnydd ymarferol. Mae cydweithrediadau rhwng clinigwyr, academyddion (o gefndir epidemiolegol a gwyddonwyr data), arbenigwyr moesegol, cleifion a phobl sy’n defnyddio gwasanaethau’n hanfodol er mwyn datblygu’r offerynnau arloesol hyn.”