Built to Wake: How Hospital Noise and Light Undermine Patient Sleep

Key findings

Hospitals are measurably too loud at night. Julie Darbyshire and Duncan Young found that sound on intensive care units never dropped below 45 dBA and stayed between 52 and 59 dBA for half of the time (Darbyshire & Young, 2013, Critical Care), against a WHO night guideline near 35 dB average and 40 dB maximum. Noise reduction moves sleep in the predicted direction and dose: Mathias Basner measured sleep efficiency 4.7 percent lower in the loudest home exposure quintile than in the quietest (Basner et al., 2023, Sleep Health).

Light helps the patients who need it most. Circadian lighting that pairs bright days with dark nights improved sleep duration and sleep onset in dementia patients (Figueiro et al., 2023, Frontiers in Physiology) and in psychiatric inpatients (Volf et al., 2025, Chronobiology International). Engineered dynamic lighting for healthy adults does not reproduce that benefit (Rongpeng Zhang et al., 2020, International Journal of Environmental Research and Public Health; Collier et al., 2023, Buildings).

By design, the certainty here is uneven. The direction and dose of both levers rest on evidence that replicates across clinical trials in the ward and observational work in the home, which is why I rate this confident on sleep quality. It does not claim the step from better sleep to faster recovery, where the evidence is thin.

The order of action is clear. Cutting night noise toward the WHO floor and building genuine day-night light contrast are the two levers with robust, low-cost support. Specifying complex tunable lighting as a sleep measure for general wards is not yet warranted.

Hospitals run loud and bright around the clock

The unit of analysis here is sleep quality, not recovery, and the environment acts on it through two main channels: noise and light. Both are chronically elevated in the places patients are meant to heal.

On noise, the measurements are consistent and unflattering. Darbyshire and Young recorded sound on intensive care units that always exceeded 45 dBA and sat between 52 and 59 dBA for half of the recording time (Darbyshire & Young, 2013, Critical Care). The WHO night guidance is roughly 35 dB on average with a 40 dB maximum, so the typical ICU runs ten to twenty decibels hot, a difference the ear registers as several times louder. This is not a quiet environment with occasional alarms. It is a loud environment throughout the night.

On light, the problem is the mirror image: too little signal by day and too much by night. The clinical lighting literature treats the relevant input as contrast, bright days against dark nights, rather than absolute level. Much of the supporting light work was done in offices and homes rather than wards, which matters for how far it transfers and is taken up in the research context below.

Noise is the better proven lever, and it is proven in the ward

The strongest single dose-response reading comes from the home. Basner found sleep efficiency 4.7 percent lower in the highest noise exposure quintile than in the lowest, at high significance (Basner et al., 2023, Sleep Health), with smaller but real decrements alongside for carbon dioxide, temperature and particulate matter. Yinxian Chen read the same direction in African American adults at home, with each step of noise exposure associated with lower sleep efficiency (β = -1.81 percent, 95 percent CI -2.00 to -0.45) (Chen et al., 2026, Sleep Health). In a controlled field experiment, Frank Schmidt found that raising the number of nighttime aircraft noise events worsened sleep quality at high significance (Schmidt et al., 2013, European Heart Journal). Basner and Chen used different samples and different methods, and both point the same way: higher noise, lower sleep efficiency.

The mechanism shows up in arousals and sleep architecture, and here the studies are clinical. Sandra Horsten pooled controlled work and found the number of arousals differed sharply between a baseline and an ICU noise condition, a mean difference of 9.59 arousals (95 percent CI 2.48 to 16.70) (Horsten et al., 2017, British Journal of Anaesthesia). Rong-fang Hu exposed healthy subjects to a simulated ICU of combined noise and light and recorded poorer perceived sleep, more light sleep, longer REM latency and less REM sleep, with melatonin moving as the mediator (Hu et al., 2010, Critical Care).

The interventions that mask or remove noise are what put this into practice, and they are cheap. Emine Arık ran a randomised trial in postoperative neurosurgical patients and reported far better sleep in the intervention arm than in controls, a Richards-Campbell score of 80.61 against 33.50 (Arık et al., 2020, World Neurosurgery). Evansaralin Warjri found a white-noise app improved sleep on each of three days in critically ill patients (Warjri et al., 2021, Nursing in Critical Care). Bernard Ong cut the average sound level on an intensive care unit by 4.8 dB through a behaviour and training intervention (Ong et al., 2025, World Journal of Critical Care Medicine). For darkness and quiet combined, Karina da Silva Higa pooled ICU trials of earplugs and eye masks and found improved subjective sleep quality at moderate certainty, with eye masks against usual care giving a mean difference of about 10 points on a 100-point scale (95 percent CI 7.97 to 12.03) (da Silva Higa et al., 2024, Nursing in Critical Care).

The counter-voice on noise

The contradicting reading belongs to Katherine Thomas, and it should be named. Thomas reported that on her wards night noise was already adequately low at 35 to 40 dB and was not improved further by sleep-promoting interventions (Thomas et al., 2012, Journal of Hospital Medicine). Thomas does not contradict the dose-response work, she marks its limit: once a ward already sits near 35 dB, an intervention has little room to move it. Basner and Chen still hold above that level.

Position

The noise-to-sleep link is robust in direction and dose, and the intervention evidence is strongest exactly where the problem is worst, the intensive care unit. But it rests heavily on subjective sleep scores. da Silva Higa is explicit that earplugs and eye masks improved subjective sleep quality while objective sleep measures moved less (da Silva Higa et al., 2024, Nursing in Critical Care). So I read masking and blocking as reliably effective on perceived sleep, with thinner support on measured sleep architecture. For a lever this cheap, that is still the best ratio of evidence to cost in the field.

Light helps the patients who need it most, not everyone

On the light side the direction is consistent across many groups, but the strength of the evidence splits along a line that matters for design. In vulnerable inpatient populations, circadian lighting works. Mariana Figueiro showed that lighting delivering bright days and dark nights improved sleep duration and sleep onset in patients with moderate to severe dementia (Figueiro et al., 2023, Frontiers in Physiology). Carsten Volf found the same pattern in patients with major depression on a psychiatric ward, with a dynamic scenario peaking at 576 melanopic lux by day against a static 66 lux improving sleep duration, waking frequency and sleep end (Volf et al., 2025, Chronobiology International). Gunnhild Hjetland reported significant proxy-rated sleep improvement in dementia patients by week 16 and week 24 of bright light treatment, though the gain did not appear in actigraphy (Hjetland et al., 2021, BMC Geriatrics). The clinical signal is real, and it is concentrated in the patients with the most disrupted circadian systems.

The daylight evidence points the same way and adds a dose. Mohamed Boubekri found office workers in an optimised daylight condition, 316 against 40.6 equivalent melanopic lux, slept 37 minutes longer on actigraphy, with the gain largest for the worst baseline sleepers (Boubekri et al., 2020, International Journal of Environmental Research and Public Health). Angus Burns found each hour of daytime light associated with fewer insomnia symptoms and an earlier chronotype in a large cohort (Burns et al., 2021, Journal of Affective Disorders).

The counter-voice on engineered dynamic lighting

Here the evidence genuinely splits. Rongpeng Zhang reported a significant decrease in perceived sleep quality and sleep time after office occupants were exposed to a dynamic lighting schedule (Rongpeng Zhang et al., 2020, International Journal of Environmental Research and Public Health). Jessica Collier found no significant differences on any measure and read this as a sign of how hard realistic in-situ lighting studies are (Collier et al., 2023, Buildings). Isabel Schöllhorn found no evidence that an eight-hour scenario with static or dynamic virtual-sky lighting affected alertness, cognitive performance or morning cortisol against standard lighting in healthy young men (Schöllhorn et al., 2023, PLoS ONE).

The reconciling reading is that timing and contrast drive the effect, not the dynamic schedule as such. Boubekri’s contrast worked where the engineered schedules in healthy offices did not. M. E. Kompier is the named voice that timing and dynamics, not nominal level, are what move the outcome (Kompier et al., 2022, Building and Environment).

Position

The benefit of daylight and of bright-day against dark-night contrast is robust, and it is strongest in clinical groups and in people with poor baseline sleep. The benefit of engineered, tunable dynamic lighting in healthy general populations is not established. The vendors have run ahead of that distinction. A dementia ward or a depression unit has the evidence for a circadian lighting specification. A general ward, occupied by people with working circadian systems, does not. There the cheaper move wins: bright daylight and genuinely dark nights.

Air, humidity and temperature point the right way but are too thin to decide on

The third cluster is present in the data but thinner, and the report treats it as secondary on purpose. Xinbo Xu found sleep quality fell as carbon dioxide rose, with the questionnaire score at 3000 ppm only 80.8 percent of the score at 800 ppm (Xu et al., 2020, Indoor Air). Jatuporn Wanchaitanawong, working across a very large nightly dataset, found each one-point rise in relative humidity reduced REM proportion by 0.064 percent and deep-sleep proportion by 0.081 percent (Wanchaitanawong et al., 2026, Sleep). The temperature evidence is weakest of all: L. Lan reported mean sleep efficiency rising from 84.6 percent without cooling to 95.3 percent with combined back and head cooling at 32 degrees (Lan et al., 2018, Indoor Air), which is one laboratory result rather than a body of work.

The counter-voice here disciplines the whole cluster. Chen found that a combined index of all environmental factors was associated with lower sleep efficiency at a level that was not significant (β = -1.38 percent, 95 percent CI -3.01 to 0.26) (Chen et al., 2026, Sleep Health). Read alongside the single-factor findings, that null on the combined index warns against assuming these inputs act independently and additively in a real bedroom.

What the evidence does not yet show

The noise and daylight readings are robust in direction and each has at least one clean dose-response anchor, Basner for noise and Boubekri for daylight. The clinical lighting work in dementia and psychiatric populations is the most internally consistent part of the light evidence. Three gaps remain, and the report names them rather than writing over them.

First, the link this report is built around is sleep quality, not recovery. The corpus assembled here does not show the step from better sleep to faster healing. A few studies touch physiological markers, melatonin and inflammation after abdominal surgery (Yaşar et al., 2017, Molecules) and after craniotomy (Arık et al., 2020, World Neurosurgery), and Marie Engwall measured self-reported recovery after an environmental intervention (Engwall et al., 2021, HERD), but none establishes the causal chain from environment through sleep to clinical outcome. That chain is plausible and worth a dedicated report. It is not proven here.

Second, much of the lighting evidence comes from offices and homes rather than wards, and its transfer to the inpatient setting has not been tested systematically. The clinical lighting trials that do exist sit in dementia and psychiatric care, not general acute wards.

Third, none of these sources quantifies the cost of the interventions, which is where a specification decision is actually made. The evidence ranks the levers by effect. It does not rank them by cost, and a planner needs both.

Implications

For hospital planners

Design the envelope and the systems so the two robust levers are available by default rather than added later. That means night acoustic separation planned to bring recovery rooms and wards toward the WHO night floor, not a reliance on patients fitting earplugs after the building is built. It means daylight access and view treated as a sleep input, with orientation, glazing and shading that deliver bright days and genuinely dark nights, and electric lighting that supports the same contrast.

For facility and nursing leads

The cheapest interventions have the most direct evidence. Earplugs, eye masks, white-noise masking and a nighttime noise-reduction protocol improved sleep in the ICU studies and cost little. Treat them as a clinical measure with an evidence base, not as a comfort amenity. Behaviour and training reduced measured sound on a real unit by nearly 5 dB (Ong et al., 2025, World Journal of Critical Care Medicine), which is achievable without capital spend.

For procurement

Stop-doing first. Do not buy complex tunable or dynamic lighting systems as a sleep intervention for general wards on the current evidence. The clinical benefit is established for dementia and psychiatric inpatients, where a circadian specification is defensible, and not for healthy general populations, where it is contested. Where a circadian system is specified, specify it for the population the evidence covers, and put the saved budget into acoustic separation and daylight, which the evidence supports across the board.

Methodology

Verified corpus state, passing its quality check on 9 June 2026: 11,092 studies sourced, 6,247 in the topic core, 16,015 live evidence cards. This report draws on the 15 environment-to-sleep-quality factors released into the synthesis layer from the 10 June 2026 factor run, 11 rated robust, 3 preliminary and 1 speculative, each tagged to the taxonomy outcome for sleep quality.

The selection criterion: a factor entered the report only if it linked an environmental stimulus to sleep quality on data-bearing sources, meaning measured studies, syntheses or simulations. Guidelines and reviews provided context rather than the single number. Opinion sources appear only as named counter-voices.

Two independent sampling logics back the central claims: randomised and quasi-experimental trials, largely clinical (Arık, Volf, Figueiro, Hjetland, Ong), and observational or cohort work in the field (Basner, Chen, Schmidt). At least one named counter-voice sits on each lever, Thomas on noise and Rongpeng Zhang, Collier and Schöllhorn on dynamic lighting.

Limits of the evidence: the sleep-to-recovery link is not established in this corpus, the lighting evidence transfers from non-clinical settings that have not been tested on general wards, and no source quantifies intervention cost. The report is rated confident on sleep quality and explicitly not on recovery.