The AI Journal

In September 1952, Virginia Apgar, an anesthesiologist at Columbia University’s Presbyterian Hospital, presented a paper at a conference of anesthesia researchers describing a problem she had grown tired of watching go unmeasured. Obstetric literature on infant resuscitation was, in her own words, full of imaginative ideas and unscientific observation, with no consistent way to compare one delivery room practice against another. Her solution, published the following year as A Proposal for a New Method of Evaluation of the Newborn Infant, assigned a baby a score from zero to ten based on five simple signs, checked at one minute after birth and again at five, so that different anesthetics and different resuscitation techniques could finally be compared on the same scale. It was a tool built for research audit, not for diagnosis, and certainly not for predicting what a particular baby’s brain would look like at age six. Seventy years later, a low Apgar score routinely appears in courtrooms as evidence that a specific child’s cerebral palsy was caused by a specific failure during a specific delivery. The tool has outgrown the question it was built to answer.

The score itself has barely changed since 1953: heart rate, respiratory effort, reflex irritability, muscle tone and colour, each worth up to two points. What has changed is the backronym. Apgar’s own paper used no acronym at all. In 1961 a Denver resident working under the pediatrician L. Joseph Butterfield noticed that the five criteria could be rearranged to spell her name, and by the time the mnemonic Appearance, Pulse, Grimace, Activity, Respiration had spread through American medical training, the score had acquired a kind of definitional solidity it never had for its inventor. Apgar had proposed five practical bedside signs a delivery-room nurse could learn in an afternoon. What circulated afterward looked, to anyone encountering it fresh, like a validated diagnostic instrument with the same claim to authority as a laboratory test.

That appearance of authority is precisely what the American College of Obstetricians and Gynecologists and the American Academy of Pediatrics moved to correct in a joint committee opinion, most recently revised in 2015. The statement is unusually direct for a professional consensus document. The Apgar score, it says, cannot be considered evidence of or a consequence of asphyxia, does not predict individual neonatal mortality or neurologic outcome, and should not be used for that purpose. It adds that an Apgar score assigned during active resuscitation is not equivalent to one assigned to a baby breathing spontaneously, since the very interventions that save the infant also depress the numbers meant to measure how much saving was needed. The two bodies most responsible for how American obstetric medicine actually practises felt it necessary to state, in a formal opinion, that a number nearly every clinician in the country records on nearly every birth should not be used the way many of them had been using it.

The gap between design and use did not appear by accident. Apgar’s five-minute score was meant to answer a narrow question about a delivery just concluded: how vigorously, on average, do babies respond to this anesthetic compared with that one. It was never validated as a test of an individual child’s long-term neurological trajectory, because that was not the question it was built to answer, and a single snapshot of heart rate and colour taken minutes after birth was never going to carry that kind of evidentiary weight for one particular child. But the score has three properties that make it almost impossible to keep confined to its original purpose. It is recorded on essentially every birth in the developed world, it is expressed as a single number that looks quantitative and therefore objective, and it is entered into the medical record at the exact moment, immediately after birth, that any later injury will eventually be traced back to. Litigation does not need a score that was validated for individual prediction. It needs a score that exists, that looks precise, and that was written down before anyone knew whether there would be a lawsuit. The Apgar score satisfies all three conditions perfectly and the first not at all.

Birth injury law firms now routinely explain to prospective clients that a low Apgar score can help establish that a child’s cerebral palsy stemmed from a birth complication, and that consistently low scores across repeated checks may indicate a preventable injury. This is not a fringe misreading confined to advertising copy. Cerebral palsy litigation has produced multi-million-dollar verdicts substantially anchored to Apgar scores of zero recorded at birth, and defense experts spend as much time arguing that a good score means an injury happened later as plaintiff’s experts spend arguing that a bad one proves it happened during delivery. Both sides are extracting individual causal conclusions from a score whose authors, and the two largest professional bodies overseeing its use, say cannot bear that weight.

The strongest objection to treating this as pure overreach is that the score is not actually useless for the questions litigation asks of it; it is only useless for asking them about one baby at a time. Population studies really do show that a five-minute Apgar score of five or below is statistically associated with a materially higher risk of cerebral palsy and of neonatal mortality than a reassuring score of seven to ten. The American Academy of Pediatrics’ own 2014 report on neonatal encephalopathy accepts this and treats a persistently low score as one legitimate flag among several, alongside umbilical cord blood gas values and placental pathology, that something may deserve closer investigation. If a population-level association is real and clinically accepted, dismissing every courtroom use of the score as junk science overstates the case.

This concedes something real without rescuing the practice as it is actually conducted. A statistical association across thousands of births licenses a probabilistic statement about a population; it does not license the individual causal claim that a jury is being asked to accept about one child. The professional guidance is explicit that the same low score can result from prematurity, from congenital anomaly, from a healthy infant simply slow to transition, or from the resuscitation itself, and that distinguishing between these explanations requires exactly the additional evidence, cord gases, placental examination, a full obstetric history, that Apgar’s one-page bedside score was never designed to supply. Using the score as one data point within that fuller workup is defensible and is what the guidelines actually recommend. Using it as the anchor of an individual causation argument, the way both plaintiff and defense routinely do, treats a population correlation as though it settled a single case, which is a different and considerably less defensible move.

Apgar died in 1974, twelve years before the first published committee opinion attempting to rein in the score’s use as a predictor rather than a snapshot. She built an instrument to answer a question about anesthesia practice that was cluttering the obstetric literature with unmeasured impressions, and she built it to be simple enough that any nurse could apply it under delivery-room pressure. Simplicity and ubiquity turned out to be exactly the properties that made the score portable to a use she never intended and her professional successors have spent decades trying to disclaim. The number survived the test of time. The question it was built to answer did not survive the number.