The psychiatric inpatient unit at the University of Alberta in Edmonton had a quirk that nobody had thought of as an experiment. Its rooms were positioned so that half were bright and sunny and the rest were not.
Kathleen Beauchemin and Peter Hays went back through the records of severely depressed patients admitted to that unit and compared how long each group stayed. The people in the sunny rooms went home after an average of 16.9 days. The ones in the dull rooms stayed 19.5. A gap of 2.6 days, about fifteen per cent, at p < 0.05. Their paper ran to three pages in the Journal of Affective Disorders in September 1996, and its reasoning was disarmingly simple: some of these patients had been receiving light therapy without anyone deciding to give it to them.
This is one study, not settled consensus. It is also the study that opened a line of enquiry now three decades old, one that has since produced retrospective agreement in two other countries and three randomised trials that failed to reproduce the effect. That whole arc is more interesting than the number it started with.
What the paper actually compared
Beauchemin and Hays did not run a trial. They looked back at case records from a ward they worked on, sorted patients by the kind of room they had occupied, and compared discharge dates. Nobody was allocating the sunny rooms to the patients most likely to improve, which is the feature that makes the comparison worth anything at all, but it falls well short of randomisation and the authors did not claim otherwise.
Length of stay also measures something looser than recovery. Discharge is a decision, made by a clinician weighing medication response, home circumstances, risk, and the pressure of the next admission waiting for a bed.
The sex split in the Edmonton data is the part quoted least and worth the most caution. The advantage was 6.8 days in men and 0.7 days in women. Francesco Benedetti’s group in Milan later offered an explanation: given how unipolar and bipolar depression are distributed between the sexes, an apparent male effect may have been a bipolar effect wearing a disguise. Their own data would go on to support that reading.
Edmonton sits at roughly 53.5 degrees north. Winter daylight there is a different quantity from winter daylight almost anywhere the paper has since been cited.
Milan, then Mallorca
Benedetti and colleagues at San Raffaele had a ward with the same accidental design and went looking. Their 2001 paper, in the same journal, covered 415 unipolar and 187 bipolar depressed inpatients in east-facing or west-facing rooms. Bipolar patients in the east rooms, which take direct morning sun, averaged 3.67 days less in hospital. In the unipolar group there was no effect at all.
So the direction agreed with Edmonton and the subgroup did not, in a way that fit the explanation Benedetti had proposed for the Canadian sex gap. His team described their own work as a naturalistic retrospective observation requiring prospective confirmation, which is more restraint than most of the citations it has since collected.
A third pass came from Mallorca. Writing in European Archives of Psychiatry and Clinical Neuroscience in 2016, Canellas and co-authors compared depressed patients across two hospitals, one of which admitted roughly three times the accumulated light of the other. Patients in the darker building stayed a median 14 days, those in the brighter one 11, a reduction that held for the group as a whole and vanished in the subgroups.
Canada, Italy and Spain, pointing the same way each time, and in every case the allocation was whatever the hospital happened to be doing anyway.
What happened when someone randomised it
Modified hospital lighting has since been tested against length of admission in three randomised trials. No reduction in any of them.
Niels Okkels and a team in Aarhus allocated 54 inpatients to a pre-set circadian lighting environment or standard lighting. Their primary outcome was sleep quality on the Pittsburgh Sleep Quality Index, which improved slightly more in the intervention group without reaching significance. Length of stay came out at 22 days in the lit ward and 19 in the control group. Markus Canazei’s group, reporting in Scientific Reports in 2022, tracked 30 inpatients with depressive disorders in rooms with dynamic or standard bedroom lighting, and found improvements in several sleep and circadian measures. Admission lasted 20 days on one side and 21 on the other.
Norway designed the difference into a building.
St Olavs Hospital in Trondheim finished construction of a 40-bed acute psychiatric unit in December 2017, divided into two wards with identical layouts, facilities and staffing levels, differing only in the evening light spectrum. Staff rotated between them every six weeks. Håvard Kallestad’s team randomised 476 patients across the two wards and published the result in PLOS Medicine in December 2024. Mean duration of admission was 7.1 days in the blue-depleted ward and 6.7 in the standard ward, an estimated difference of 0.4 days, p = 0.523.
The authors note in their discussion that theirs is the third such trial to miss on length of admission, and they give the reason plainly: discharge timing depends on vacancies at other facilities and on the need to free the bed for the next acute admission, which blunts the measure considerably. Their secondary outcomes did move. Clinicians rated patients in the blue-depleted ward as more improved at discharge and less severely unwell, and recorded lower levels of aggressive behaviour, with numbers needed to treat between seven and twelve.
Where the evidence for light itself still stands
The case for light itself survives all of this intact. A team led by Raymond Lam ran a randomised, double-blind, placebo-controlled trial reported in JAMA Psychiatry in 2016, enrolling 122 adults with non-seasonal major depressive disorder across four arms over eight weeks. On the primary outcome, change in Montgomery-Åsberg Depression Rating Scale score at week eight, light monotherapy separated from placebo and the light-plus-fluoxetine combination separated further. Fluoxetine alone did not separate from placebo in this trial.
That last sentence needs its scope attached before anyone runs off with it. One trial, 122 people, eight weeks, a single drug at a single dose, one rating scale, one outpatient population, conducted in southern Canada. That is a long way from establishing light as the better option in general, and the trial was never designed to ask.
There is also a mismatch running through the whole comparison. Beauchemin, Benedetti and Canellas were describing morning daylight arriving through a window. Okkels, Canazei and Kallestad tested indoor lighting systems, two of them focused on removing blue wavelengths in the evening. Kallestad’s team say as much, listing morning bright light for non-manic patients among the things future trials should add.
Sixteen sunrises, none of them usable
Crew aboard the International Space Station pass through roughly sixteen sunrises and sunsets every 24 hours, which is functionally the same as having none. Light there has to be built rather than admitted.
From 2016, NASA began replacing the fluorescent General Luminaire Assemblies with Solid-State Lighting Assemblies, tunable LED units with settings meant to support alertness at one end of the crew day and sleep at the other. The accompanying Lighting Effects investigation, run with George Brainard at Thomas Jefferson University and Steven Lockley at Brigham and Women’s Hospital, was set up to test whether the new fixtures did anything measurable for circadian regulation, sleep and performance in flight.
That work concerns healthy crew and should not be read across to depression treatment. What it shows is how differently the variable gets treated depending on where you are standing. On a spacecraft, light is a systems problem with a specification, a budget line and a principal investigator attached to it.
What a room can change and what it cannot
The tempting misreading is the one where the window becomes the treatment. Every patient in every one of these studies was receiving standard care, and 2.6 days was a group average that says nothing about what any particular person’s recovery would have looked like across the corridor.
A review led by Jan Scott in BJPsych Advances traced the whole line back to those early ward observations and described the Trondheim unit before its trial had reported. Read now, it works as a record of a hypothesis being taken seriously enough to pour concrete for, then tested, then not confirmed on the measure everyone had been quoting since 1996.
For anyone reading this while genuinely unwell, the sensible route is the ordinary one. Whether light has any place alongside whatever else is being tried is a conversation for a GP or treating clinician, and nothing here is a reason to change or delay treatment.
Morning sun through an east-facing window, the thing Beauchemin and Hays stumbled into, is the specific intervention the Trondheim group now say should be tested next. Thirty years on, it still has not been randomised.
