The 4 p.m. Hazard: When Bad Software Becomes a Clinical Risk

By late afternoon the ward is quieter, but the day has not gotten any lighter. A physician who reasoned carefully through a difficult case at eight in the morning now faces a routine consultation, and something has shifted. Not knowledge — she knows exactly as much as she did at breakfast. What has drained away is the willingness to do the slow, effortful thinking that the morning came so easily by.

European medicine has spent decades regulating the body. We cap shift lengths and protect rest periods because we accept that an exhausted physician is a hazard to patients. We have built almost nothing around the exhaustion that leaves no mark on a roster: the steady erosion of judgement over the course of a shift.

Two kinds of thinking, one finite tank

Daniel Kahneman's distinction is by now familiar. There is fast thinking — reflexive, automatic, almost free — and there is slow thinking: deliberate, analytical, effortful. Complex diagnosis and the harder conversations with patients draw on the slow kind. Routine, the fast kind.

The slow kind appears to run on a finite reserve, and that reserve depletes through use. This is where the design of clinical software stops being a convenience question and becomes a clinical one. Every clumsy menu, every password reset, every alert that fires for no reason and must be read and dismissed, spends a small unit of the same attention a clinician will later need at the bedside. We have built systems that tax the most expensive resource in the building to perform the cheapest tasks in it.

The path of least resistance

When the reserve runs low, the mind does not stop deciding. It changes how it decides. It reaches for the fast, low-effort option — the path of least resistance — and that path is not always the right one.

A much-cited study of primary-care clinics found that as a session of consultations wore on, doctors became measurably more likely to prescribe antibiotics for conditions that did not warrant them. The reading is uncomfortable and plausible. Explaining to an anxious patient why they do not need a prescription is effortful work. Writing the prescription is mechanically easy. A depleted clinician, all else equal, drifts toward the easy thing. The knowledge was intact; the energy to act on it was not.

Notice what this does to where we look for error. The mistake at four o'clock did not begin at four o'clock. It began in the hundred small frictions of the hours before it.

Attention as protective equipment

We give radiologists lead aprons and surgeons sterile gloves without calling either a perk. They are protective equipment against an environmental hazard. It is worth treating a clinician's attention with the same seriousness. Software that pre-fills the standard form, that suppresses the nine alerts in ten that no one needed to see, that handles the billing codes without a human in the loop, is not a comfort feature. It is shielding the one faculty you most need intact at the end of a long day.

There is a practical test in this for anyone who reviews adverse events. When you analyse a clinical error, do not stop at the moment of failure. Look at the hour before it. Count the unnecessary clicks, the redundant alerts, the micro-decisions the clinician had to clear before they reached the one that mattered. If the tools had already spent the attention the patient needed, the tools are part of the cause — and they are the part you can actually fix.