Eight ways in which AI is transforming healthcare

AI has been at the fore­front of the med­ical profession’s efforts to fight Covid-19 and treat patients dur­ing the coro­n­avirus pan­dem­ic. Enabling health­care providers to make fast, accu­rate and data-dri­ven deci­sions, the tech­nol­o­gy has been pro­duc­ing some extra­or­di­nary out­comes.

Out­side the Covid cri­sis, machine intel­li­gence is lend­ing itself to hun­dreds of med­ical appli­ca­tions, from scan­ning vast num­bers of peo­ple to assess their risk of demen­tia to accel­er­at­ing the drug dis­cov­ery process. Here is just a small selec­tion of cas­es where the tech­nol­o­gy is rev­o­lu­tion­is­ing health­care pro­vi­sion.

1. Easing the admin burden

Health­care pro­fes­sion­als are using AI-pow­ered speech-recog­ni­tion sys­tems to update elec­tron­ic patient records more quick­ly and accu­rate­ly.

“This has had a big impact on the effi­cien­cy of get­ting our let­ters done for any clin­ic,” reports Dr Paul Alt­mann, chief clin­i­cal infor­ma­tion offi­cer at Oxford Uni­ver­si­ty Hos­pi­tals NHS Foun­da­tion Trust. “Clin­i­cians can now send let­ters with­in 24 hours, or even instant­ly if they aren’t wait­ing on blood test results. Before the speech-recog­ni­tion era, let­ters took two weeks or even longer to be com­plet­ed.”

Dr Simon Wal­lace is chief clin­i­cal infor­ma­tion offi­cer at Nuance, a provider of speech-recog­ni­tion sys­tems. He says that clin­i­cians spend an aver­age of 11 hours each week cre­at­ing doc­u­ments. If you fac­tor in time wast­ed on lost and repeat­ed doc­u­men­ta­tion, they could be spend­ing up to half of their work­ing week on admin – a lot of time to be away from their patients.

“We speak three times faster than we type. By using AI-pow­ered speech recog­ni­tion, clin­i­cians save time on doc­u­men­ta­tion process­es, which enables them to spend more of it on deliv­er­ing care,” he says. “Even when clin­i­cians talk extreme­ly quick­ly at length, speech-recog­ni­tion tech­nol­o­gy can recog­nise and record all their words, trans­form­ing them into detailed med­ical notes. Thanks to the capa­bil­i­ties of the soft­ware, all med­ical terms are auto­mat­i­cal­ly recog­nised in the cor­rect con­text. The only thing that users need to do is use their voice.”

Wal­lace believes that this facil­i­ty could sig­nif­i­cant­ly reduce the admin bur­den asso­ci­at­ed with remote con­sul­ta­tions and could even help to reduce the risk of burn-out among NHS staff, many of whom have been stretched to the lim­it by the Covid cri­sis.

2. Improving communications with patients

Con­vey­ing infor­ma­tion to the pub­lic in a time­ly and effec­tive way has been cru­cial dur­ing the Covid-19 vac­ci­na­tion pro­gramme. It’s there­fore vital that they don’t ignore impor­tant mes­sages. This was the chal­lenge faced by phar­ma­cy chain Wal­greens as a vac­cine provider in the US.

Bri­an Tyrrell, its senior direc­tor respon­si­ble for cus­tomer mar­ket­ing plat­forms, stress­es the impor­tance of the word­ing of the company’s mes­sag­ing. “You’d expect peo­ple to be more respon­sive to an email telling them crit­i­cal health infor­ma­tion, of course, but we had a duty to ensure that our mes­sages were heard,” he says.

Wal­greens, which can reach about 50 mil­lion peo­ple via email, has been using Phrasee’s AI tech­nol­o­gy. The sys­tem learns with each com­mu­ni­ca­tion the type of lan­guage that con­nects best with the tar­get audi­ence. “This means that we can dou­ble down on our efforts to reach the max­i­mum num­ber of peo­ple with rel­e­vant updates as time moves on,” Tyrrell says.

Wal­greens esti­mates that 30% more of its email recip­i­ents than nor­mal have opened a mes­sage from the com­pa­ny since it start­ed apply­ing the Phrasee sys­tem to the vac­ci­na­tion cam­paign. Poten­tial­ly, that could mean up to 30% more peo­ple than nor­mal read­ing infor­ma­tion about appoint­ments – and up to 30% more get­ting vac­ci­nat­ed.

3. Monitoring for adverse events

Agen­cies such as the US Food and Drug Admin­is­tra­tion and the UK Med­i­cines and Health­care Prod­ucts Reg­u­la­to­ry Agency have sys­tems in place to mon­i­tor reac­tions to any drug, device or ther­a­py. Bio­phar­ma com­pa­nies also have their own sys­tems for mon­i­tor­ing adverse events for their prod­ucts. These are under more pres­sure than nor­mal owing to the Covid cri­sis.

Dr Vladimir Makarov is con­sul­tant and pro­gramme man­ag­er at the AI Cen­tre of Excel­lence at Pis­toia Alliance, a not-for-prof­it body advo­cat­ing greater col­lab­o­ra­tion in health­care and life sci­ences. He says that AI sup­ports the mon­i­tor­ing process, “which will be strain­ing under the extra work­load imposed by the world­wide vac­cine roll-out. AI also helps to inter­ro­gate data and evi­dence from var­i­ous health­care envi­ron­ments, from under­stand­ing the effec­tive­ness of a drug in a real-world set­ting to analysing genome sequenc­ing.”

AI is also being used to iden­ti­fy exist­ing treat­ments that can be repur­posed for patients. It gives researchers an effi­cient way of review­ing large vol­umes of data and uncov­er­ing new insights. For instance, the Pis­toia Alliance recent­ly col­lab­o­rat­ed with a group of part­ners on an AI ‘datathon’ for drug repur­pos­ing. In one month, this helped sci­en­tists to dis­cov­er four repur­pos­ing can­di­dates with poten­tial to treat chron­ic pan­cre­ati­tis – a dis­ease affect­ing about 1 mil­lion peo­ple world­wide – which cur­rent­ly doesn’t have an approved treat­ment.

4. Automating Covid-19 reports

The pan­dem­ic has pre­sent­ed many new chal­lenges, one of which has been the intro­duc­tion of addi­tion­al dai­ly report­ing respon­si­bil­i­ties.

As part of the NHSX AI Lab project, tech con­sul­tan­cy Foundry4 took on the task of automat­ing the dai­ly Covid sit­u­a­tion reports being sent to NHS Eng­land by Ket­ter­ing Gen­er­al Hos­pi­tal NHS Foun­da­tion Trust. 

“A lot of our time was tak­en up with con­vey­ing an accu­rate pic­ture of how the hos­pi­tal was cop­ing,” says Ian Rod­dis, the trust’s act­ing dig­i­tal direc­tor. “Although it was a vital part of the nation­al response, we were adamant that the process would not pre­vent us from ded­i­cat­ing as much of our time as pos­si­ble to patient care.”

He and his col­leagues were aid­ed in their efforts by the intro­duc­tion of the Mary Bot, named after the trust’s head of IT. This is soft­ware that autonomous­ly draws data from patient and HR admin sys­tems to help man­age the report­ing process.

“Remov­ing the heavy admin­is­tra­tive bur­den from our clin­i­cal staff has not only enabled them to spend more time attend­ing to patients; it has also erad­i­cat­ed errors in our report­ing,” Rod­dis says. “It’s ensured that we’re pro­vid­ing an accu­rate pic­ture that sup­ports that nation­al pan­dem­ic response.”

On top of this, the automa­tion has saved an esti­mat­ed 4,400 work­ing hours a year, rep­re­sent­ing an annu­al cost sav­ing of more than £150,000. 

5. Reducing stroke recovery times

Stroke sur­vivors are often left with long-term health prob­lems such as impaired mobil­i­ty, which can require peri­ods of reha­bil­i­ta­tion last­ing months or even years. Few­er than a third of patients ful­ly regain their mobil­i­ty and strength.

ReLive, a start-up based at Nazarbayev Uni­ver­si­ty in Nur-Sul­tan, Kaza­khstan, is work­ing to improve stroke patients’ recov­ery times with the aid of reha­bil­i­ta­tion robots. These devices are wear­able exoskele­tons that can be strapped around the affect­ed limb. They are con­trolled using soft­ware that reads and inter­prets the patient’s brain sig­nals via an elec­troen­cephalo­gram (EEG). Infor­ma­tion about their progress can be record­ed dur­ing the reha­bil­i­ta­tion exer­cis­es on an online plat­form for clin­i­cians to analyse.

The sys­tem can inter­pret the sig­nals obtained by biosen­sors, such as an EEG, and pass these on to the robot con­troller. It can also learn an individual’s gait to fine-tune the robot’s move­ments and opti­mise these so that the stress on the soft tis­sues sur­round­ing the affect­ed limb is min­imised.

6. Detecting the early warning signs of dementia

AI-enabled sys­tems have come to the fore­front in the diag­no­sis of Alzheimer’s dis­ease and oth­er forms of demen­tia. For instance, they can note slight changes in speech, such as the elon­ga­tion of paus­es between words, a grow­ing pref­er­ence for pro­nouns rather than prop­er nouns and the use of over­ly sim­plis­tic descrip­tions.

View­Mind is one com­pa­ny that’s using AI to detect Alzheimer’s and sim­i­lar con­di­tions at the pre-clin­i­cal stage. Using a VR-type head­set in a test that takes about 10 min­utes, it’s aim­ing to assess 1 bil­lion peo­ple a year in set­tings such as GP surg­eries and opti­cians as part of a stan­dard health check. The sys­tem cap­tures more than 10,000 data points through eye move­ments. The com­pa­ny says that this can detect cog­ni­tive prob­lems 20 years before the symp­toms present them­selves.

View­Mind recent­ly con­duct­ed a study that has found a sig­nif­i­cant por­tion of Covid-19 patients have a form of so-called long Covid that could trig­ger poten­tial­ly seri­ous cere­bral side effects. The researchers used the head­set device on patients to stim­u­late parts of their brains and mea­sure their eye respons­es. The results were fed through an AI sys­tem to analyse the cog­ni­tive impact of the virus.

7. Improving fertility treatment

Vir­tu­al fer­til­i­ty clin­ic Apric­i­ty uses a pro­pri­etary algo­rithm to pre­dict its patients’ chances of suc­cess­ful con­cep­tion, with and with­out treat­ment. 

“Our patients have found great val­ue in using our AI fer­til­i­ty pre­dic­tor tool, which analy­ses lifestyle fac­tors to eval­u­ate the like­li­hood of preg­nan­cy,” says Dr Cristi­na Hick­man, Apricity’s chief sci­en­tif­ic offi­cer. “It can sug­gest dif­fer­ent types of treat­ment to sup­port them in their quest for fer­til­i­ty.”

She con­tin­ues: “Our team is devel­op­ing machine-learn­ing mod­els with appli­ca­tions in a num­ber of areas, includ­ing embryo selec­tion, fol­li­cle-scan inter­pre­ta­tion and hor­mon­al treat­ment. When we’re cul­tur­ing embryos, for instance, we can cap­ture many images and mil­lions of data points. We can use these to cre­ate an algo­rithm-based stan­dard clas­si­fi­ca­tion, where we assign a score to each embryo and pre­dict its chances of becom­ing a healthy baby. Com­pared with tra­di­tion­al meth­ods, this would save the time, mon­ey and stress usu­al­ly spent on unvi­able embryos, result­ing in more babies per patient and bet­ter stan­dards of clin­i­cal prac­tice.”

AI can also assist the clin­i­cians in stan­dar­d­is­ing deci­sions such as choos­ing the best sperm and deter­min­ing the opti­mal num­ber of embryos for trans­fer to max­imise suc­cess rates while min­imis­ing mul­ti­ple births.

“Ulti­mate­ly, AI can help to involve patients in the key deci­sions,” Hick­man says. “This brings sim­plic­i­ty, auton­o­my and less stress to a jour­ney that, with­out AI, is any­thing but straight­for­ward, con­trol­lable and stress-free.”  

8. Accelerating pharmaceutical R&D

Machine learn­ing is prov­ing invalu­able to the phar­ma indus­try in the drug dis­cov­ery process and the test­ing of new treat­ments. Giv­en that bring­ing a new drug to mar­ket can take up to 15 years and be an extreme­ly cost­ly and risky exer­cise, many phar­ma­ceu­ti­cal com­pa­nies are devot­ing more resources to find­ing new uses for exist­ing prod­ucts.

“Even though machine learn­ing is used at all stages of drug dis­cov­ery, it is of most use at the start, when there is a large amount of data to process and con­nect,” says Aman­da Schierz, prin­ci­pal data sci­en­tist at AI plat­form DataRo­bot. “This is the best stage to pre­dict any prob­lems that could occur fur­ther down the line. Towards the end of the process, machine learn­ing is used pre­dom­i­nant­ly as a sup­port tool for human deci­sion-mak­ers, enabling researchers to inves­ti­gate alter­na­tive com­bi­na­tions and can­di­dates quick­ly.”

She observes that the speed and pro­duc­tiv­i­ty of auto­mat­ed machine learn­ing align with the drug-dis­cov­ery phi­los­o­phy of ‘fail fast’. This helps to reduce the num­ber of fail­ures lat­er in the process.

“With the pub­lic focus­ing on – and becom­ing more knowl­edge­able about – the drug dis­cov­ery process in rela­tion to Covid-19, it’s time for com­pa­nies to rely more on machine learn­ing,” Schierz says. “This enables them to inno­vate and fail fast, instead of putting all their resources into one metaphor­i­cal bas­ket.”

---