Gentle sleep training cuts infant sleep problems and lifts maternal mood

The trial that measured both fears at once

In 2016 a team led by Michael Gradisar ran the experiment this whole argument needed. They enrolled 43 infants between 6 and 16 months old and randomized them to one of three arms: graduated extinction (n=14), bedtime fading (n=15), or a sleep-education control (n=14). Then they measured the two things parents actually fear alongside the sleep itself. One was salivary cortisol, the stress hormone read off a saliva swab. The other was the Strange Situation Procedure, the standard laboratory test of infant attachment, administered twelve months later.¹

Graduated extinction here means checking on a crying infant at lengthening intervals rather than leaving the room and not returning. Bedtime fading means shifting bedtime later to match when the child is actually sleepy, then walking it earlier. In both active groups, sleep latency and the number of night awakenings fell sharply. Salivary cortisol did not rise. It declined a small-to-moderate amount in both intervention arms, with no significant increase from baseline to the one-week assessment. At the twelve-month follow-up there was no significant difference between groups in secure-insecure attachment or in emotional and behavioral problems.¹

This is one small randomized trial, about fourteen infants per arm, with a single follow-up and no power to rule out a small attachment effect. Hold that limit in mind. It is still the only study that put the sleep benefit, the cortisol worry, and a gold-standard attachment measure on the same table, and every other piece of evidence gets read against it.

What "works" means, and by when

The efficacy question has the best evidence. The Melbourne Infant Sleep Study is a cluster randomized trial across 49 Maternal and Child Health centres, enrolling 328 mothers who reported an infant sleep problem at seven months. The intervention was delivered over routine nurse consultations and offered a choice of two gentle methods, not leaving a baby to cry alone. Controlled comforting means responding to the cry at increasing intervals. Camping out means sitting by the cot and fading your presence over about three weeks.²

The effect was real and modest. At 10 months, 56% of the intervention families still reported a sleep problem against 68% of controls (odds ratio 0.58, 95% CI 0.36 to 0.94, p=0.03). By 12 months the gap was 39% against 55% (odds ratio 0.50, 95% CI 0.31 to 0.80, p=0.004). Maternal depression, scored on the Edinburgh Postnatal Depression Scale where a higher number means more symptoms, ran 5.9 in the intervention group against 7.2 in controls (adjusted difference -1.7, 95% CI -2.6 to -0.7, p=0.001).² Both arms improved over the year, so some of the resolution is infants growing out of it, and the sleep problem was defined by the parent rather than by a movement sensor. This is a strong design measuring a soft outcome.

The professional grading points the same way and grades itself in the open. The American Academy of Sleep Medicine's practice parameters rate behavioral treatment overall a Standard, the strongest tier. But the specific methods a six-month parent would reach for, graduated extinction and bedtime fading, sit one rung down at Guideline, meaning moderate certainty. The Standard rating attaches to unmodified extinction, the full leave-and-do-not-return method most parents will not use. The confident headline and the graded reality diverge exactly there.³ The grading rests on a review of 52 studies, 49 of them (94%) reporting clinically significant improvement, with more than 80% of treated children improved and holding that gain for 3 to 6 months.⁴

So "works" has a precise meaning. It means parent-reported sleep and maternal mood improve over roughly three to six months, on gentle methods graded at moderate certainty. It does not yet mean anything about the child at five.⁴

The cortisol fear, stated at full strength

Here is the worry in its strongest form, the one parents have read about. In 2012 Wendy Middlemiss and colleagues studied 25 mother-infant pairs in a five-day residential program in New Zealand where nurses ran an unmodified extinction protocol. They swabbed saliva at sleep onset and again twenty minutes after the infant fell asleep, on days one and three. On day one, mother and infant cortisol moved together, correlated at r=0.78. By day three, after the infants had stopped crying at bedtime, that synchrony was gone: the mothers' cortisol dropped significantly while the infants' did not.⁵ The authors read this as a baby whose outward distress had been extinguished while the internal stress continued, no longer shared or soothed by a mother who now believed the crisis was over. In their words, the infants "were not learning how to internally manage their experiences of stress."⁵ That is a genuinely disturbing picture, and it deserves to be met on its own terms rather than waved off.

Now weigh it by its own machinery. There is no control group, so nothing says what these infants' cortisol would have done on an ordinary night. The setting is inpatient unmodified extinction in a self-referred, highly distressed population, not the home graduated extinction a typical parent uses. The cortisol sub-samples run as low as ten to nineteen infants. One co-author founds the company whose assay measured the hormone, a declared conflict.⁵ Each of these weakens the study, but one detail dissolves the headline. Read past the abstract to the numbers. Infant cortisol did not significantly change from before to after the sleep routine on day three (before M=0.512, after M=0.412, reported as not significant), and it did not differ significantly between day one and day three either.⁵

Gradisar 2016, the better design because it has a control arm, measured infant cortisol and found it did not rise and trended down.¹ Read carefully, the two studies do not even disagree about the infant numbers, because Middlemiss also found no significant change in infant cortisol. The study parents fear and the study that reassures them report the same measured thing. The disagreement is about what an unchanged cortisol level means, not about the level. That makes the fear an interpretation of a null result rather than a measured harm.¹

The long run, and the honest gap

Cortisol at three days is a proxy. The outcome parents care about is the child years later, and one trial actually looked. Price and colleagues followed up the Melbourne cohort when the children were about six years old and reassessed 225 of the original 326 families, a 69% retention rare in this field. They found no significant difference between the sleep-trained children and controls on any measure: emotional problems (p=0.8), conduct (p=0.6), the parent-child relationship for closeness (p=0.1) and conflict (p=0.4), child stress regulation, disinhibited attachment, sleep problems (9% against 7%, p=0.2), and maternal mental health (p=0.9).⁶ Their conclusion is unusually plain: "Behavioral sleep techniques have no marked long-lasting effects (positive or negative)."⁶

Notice the words in that conclusion. No lasting harm, and no lasting benefit either. By age six the control families had caught up, their children sleeping just as well, so the trial demonstrates the absence of long-term harm more than it demonstrates durable payoff. A skeptic can fairly say the early gain washes out, and that reading is true at the same time as the safety reading.⁶

The strongest in-field caution goes further, and it is worth stating. A 2016 clinical commentary concedes the efficacy record, then draws the line that matters: "there is limited evidence to conclusively support or deny that there are any lasting negative effects of extinction." No evidence of harm is not evidence of no harm. The same paper flags a plausible subgroup, warning that extinction "may exacerbate attachment issues in children for whom attachment and sleep are already impaired."⁷ Field reviews summarize the record the other way, reporting no indication of short- or long-term harm and noting that Gradisar read attachment with the Strange Situation itself rather than a questionnaire proxy.⁸ Nothing in the randomized record refutes that hypothesis, because none of these trials was powered to find a small effect in a vulnerable minority, and the guideline body itself lists long-term psychopathology as unstudied and in need of evaluation.³ On the very long term and the most vulnerable infants, the three honest words are that we do not know.

What a parent at six months does with this

The evidence sorts parents rather than dictating to them. It is strong enough to grade, and the grade is lopsided in one direction, so the decision follows from where a family actually sits.

A parent carrying a real burden, an infant waking many times a night and a mother sliding toward depression, has a genuine randomized reason to act, and a gentle method with brief checks or a faded bedtime is the version the evidence actually tested. A parent whose nights are hard but bearable can reasonably wait, because the trained and untrained children converge by age six anyway, and waiting costs the child nothing the data can find. A parent of an infant with existing attachment or regulation difficulty is in the one subgroup the trials cannot reassure, and that is a conversation with a clinician rather than a method chosen from an article.