4 February 2024

Lives are at risk when algorithms go wrong

Keeley was in agony. Why didn’t the NHS 111 software register the danger she was in?

There is an unmistakable quality to someone’s speech when they are in severe pain. Keeley had it. Something about her phrasing, short and snatched, the air sounding pent-up in her lungs as though she were trying not to exhale much as she talked. And a tremulousness to her voice that threatened any moment to dissolve her words into whimpers.

I hadn’t expected this at all. Her case had come through as a Category 3 ambulance validation. These are cases where NHS 111’s algorithmic software has recommended a paramedic crew attend the patient, but it will take hours (the official target for a “Cat 3” is 120 minutes, but it can be as long as eight hours in my area). We try to phone as many Cat 3s as we can from the GP out-of-hours service. Hardly any need an ambulance – the 111 software routinely overreacts – and we can relieve pressure on our paramedic colleagues by intervening. I’d dealt with several that evening already: a chap with headaches dating back 15 years, a student living away from home for the first time and suffering from Covid or flu, an otherwise fit 61-year-old needing painkillers for a bout of simple back pain.

Keeley’s 111 notes recorded that she might be pregnant – she was more than a week late for her period. And they detailed abdominal pain and collapse. The more I managed to gather from Keeley herself, the more concerned, and puzzled, I became. She was in a sexual relationship and, no, they weren’t using any contraception. She’d developed pain low down on the right side of her abdomen that morning that had become progressively more agonising as the day had worn on. She’d somehow managed to bathe and put to bed her two-year-old daughter but shortly afterwards the pain had come to a crescendo. She remembered suddenly feeling very lightheaded; the next thing she knew she was regaining consciousness on the upstairs landing. She thought she’d lost as much as an hour. She was lying in the exact same place: if she tried to get up she felt lightheaded and sick. Her abdomen had become noticeably swollen.

Although a little sooner than expected following the missed period, this could easily be an ectopic pregnancy – these occur in around 1 per cent of cases, where a fertilised egg implants somewhere outside the cavity of the womb. The commonest location is in a Fallopian tube. Most ectopics can be resolved without emergency care, but as the embryo grows and tries to establish a placenta, it occasionally damages and ruptures the tube, which can result in life-threatening internal bleeding. Most 111 Cat 3 cases don’t actually need an ambulance, but Keeley most certainly did – and she needed it far more quickly than 111’s algorithmic software had determined.

Keeley’s partner was away working but her mum was already en route to provide emergency childcare. I established that the paramedics would be able to gain entry to the house. Then I left Keeley to it, clutching her phone, her two-year-old sleeping soundly in her cot.

Thankfully, ambulance control answered me swiftly and upgraded her call to a Category 1. The crew were there in under ten minutes.

I’m still puzzled as to how 111’s software could have failed to recognise her case. Keeley hadn’t had any vaginal bleeding, but that’s no reassurance. The light bleeding often seen with ectopics is from the womb-lining beginning to shed as the pregnancy fails – the haemorrhage from a ruptured Fallopian tube is internal. And the 111 software hadn’t triggered an enquiry about shoulder-tip pain, something Keeley had developed since collapsing. In ectopics, this is thought to arise from blood tracking inside the abdominal cavity and irritating the diaphragm.

In 2017, 36-year-old Gail Bailey from Rotherham died from a ruptured ectopic when an incorrectly assigned Category 3 ambulance took more than two-and-a-half hours to reach her. Thankfully, Keeley’s case had a good outcome. A few days later she was doing well following emergency gynaecological surgery.