Convenience sampling was used to select patients treated in the adult ICU of a tertiary referral hospital in Nanjing from January 2023 to December 2023. The 34-bedded ICU provided specialized treatment for critical care patients across multiple specialties. The ICU comprised 22 single occupancy rooms, 2 double rooms, and 2 multi-bed rooms (totaling 8 beds), ensuring diverse environmental settings for sleep assessment. The nurse-to-patient ratio was 1:2 for mechanically ventilated patients and 1:3 for other patients. The ICU staff included consultants, residents, and nursing professionals. An existing sleep/rest guideline was implemented, focusing on maximizing patient rest and sleep by scheduling daytime rest periods and reducing nighttime activity, light, and sound levels. Clustering of care activities was a standard practice at night to minimize disruptions. Patients’ families and friends were allowed to visit from 14:00 to 20:00 each day.

The inclusion criteria were as follows: (1) The age range was between 18 and 65 years, so as to exclude the disturbance of sleep structure and comorbidities that might accompany the aging of elderly patients; (2) Conscious state clear consciousness, Glasgow Coma Scale (GCS) ≥ 14 points, to ensure that patients had clear cognitive provide informed consent; The ICU length of stay was more than 24 h to ensure that the patient was fully exposed to the ICU environment; (4) Sedative medication discontinuation time was more than 24 h, to avoid residual effects on sleep evaluation interference; ⑤ Patients voluntarily participated and provided signed informed consent.

The exclusion criteria were as follows: (1) Previous sleep disturbance including insomnia, obstructive sleep apnea syndrome, narcolepsy and other clearly diagnosed sleep-related disorders, to reduce the confounding effect of previous medical history on the study results; (2) A history of mental illness including schizophrenia, bipolar disorder, to avoid other brain states interfering with the subjective perception of sleep; (3) Communication disorders, including hearing disorders or speech disorders, so patients could provide accurate responses; (4) Patients diagnosed with delirium by Confusion Assessment Method for the ICU (CAM-ICU) before enrollment; Long-term substance dependence affecting central nervous system function or circadian rhythm; ⑥ Patients with severe pain (≥ 7 points on the numerical scale) or continuous use of analgesic drugs before enrollment. Additionally, patients who are unable to provide informed consent due to low literacy levels and have no legally authorized representatives available are excluded.

A survey was designed with six headings, namely demographic data, health data, social support, disease uncertainty, environmental factors and sleep quality. The survey items and basic information are shown in Fig. 1.

Survey items and basic information

Figure 1 presents an integrated framework for assessing perceived sleep quality in adult ICU patients, encompassing six interconnected domains. Demographic Data captures socio-contextual variables such as gender, age, and income. Health Data quantifies clinical severity via APACHE II, comorbidity burden, BMI, and documents devices, diagnosis, and pain. Social Support (PSSS) measures pre-ICU perceived support using a 12-item Likert scale. Illness Uncertainty (Mishel Scale) evaluates cognitive ambiguity. Environmental Factors record nocturnal sound and illuminance. Sleep Quality (RCSQ) assesses five dimensions via visual analog scales, with scores < 50 mm indicating significant disturbance.

The demographic data collected in the survey included health behaviours of patients, and social demographic variables such as gender, age, education level, occupation category, marital status and per capita monthly family income. Health-related data such as body mass index (BMI), Acute Physiology and Chronic Health Evaluation II (APACHEII), smoking and drinking history, and medical payment method were also collected.

The severity of disease was quantified using the acute physiological and chronic health II score, the underlying disease burden was assessed using the Charlson comorbidification index, and the pain intensity level was assessed using a numerical assessment scale (0 = no pain and 10 = the worst pain possible). In addition, the primary diagnosis, types of indwelling devices (e.g., endotracheal tubes, urinary catheters, vascular access catheters), and the frequency of nightly procedures (documented through review of nursing shift records) were recorded to identify the potential impact of treatment interventions on sleep.

The social support assessment was based on the Perceived Social Support Scale (PSSS) developed by Zimet, measuring patients’ general perceived social support preceding ICU admission. The Chinese version comprises 12 items across family, friend, and significant-other dimensions using Likert 7-point scoring. Total scores range from 12–84 (higher = stronger perceived support). While not ICU-contextualized, this instrument was selected because chronically low social support may increase vulnerability to ICU-related psychological distress that disrupts sleep through neuroendocrine pathways. The scale showed good reliability (Cronbach’s α = 0.88) in this population.

The social support assessment was based on the Perceptive Social Support Scale developed by Zimet (ref), revised to include three dimensions of family support, friend support and other support, with a total of 12 items. A Likert 7-point scale was used, with a total score ranging from 12 to 84, with a higher score indicating a stronger perceived level of social support. The reliability and validity of the scale were verified, and the internal consistency coefficient of the Chinese version was 0.88, which could effectively reflect the regulating effect of social support on the psychological state of patients.

The Chinese version of the Illness Uncertainty Scale was used to evaluate disease uncertainty, which includes two dimensions: uncertainty and complexity. The uncertainty dimension focuses on patients’ confusion about disease symptoms, diagnosis and prognosis, while the complexity dimension reflects their difficulties in understanding treatment plans and doctor-patient communication. There were 25 items in the scale, and Likert 5-level scoring method was adopted, with the total score ranging from 25 to 125 points, which were divided into three levels of uncertainty: low, medium and high. The overall internal consistency coefficient of the scale was 0.865, and the content validity index was 0.920, which was suitable for quantifying patients'cognitive uncertainty about the disease.

Environmental factors were objectively measured using calibrated instruments. Continuous monitoring was performed with a TES-1351B noise level meter (A-weighted decibels [dBA]) and a TES-1335 digital illuminance meter (Lux), positioned 30 cm above the patient's head. Sound pressure levels were recorded continuously with A-frequency weighting to reflect human auditory perception. Light intensity was measured continuously, capturing both baseline ambient levels and transient peaks (e.g., from overhead procedures). Data were logged every 60 s and downloaded hourly for analysis. The equipment was calibrated daily using certified reference standards..

Sleep quality was evaluated through the application of the Chinese version of the Richards-Campbell Sleep Questionnaire (RCSQ), initially devised by Richards for assessing sleep among non-critically ill male patients. This tool is specifically designed to assess sleep in patients with acute illness and includes five dimensions: sleep depth, sleep latency, night wake, return to sleep, and overall sleep quality. Each dimension was quantified using a 100 mm visual analogue scale in which the higher score the better the sleep quality. For this study, sleep quality was categorized as follows: sleep disturbance (≤ 49 mm), poor sleep (50–69 mm), and good sleep (≥ 70 mm). This classification is based on prior research and the specific characteristics of our study sample [1]. After reliability and validity test, the Chinese version of the scale showed that the internal consistency coefficient was 0.895 and the content validity was 0.84, which could accurately capture the fragmented characteristics of sleep in ICU environment. The disease Uncertainty scale focuses on patients'cognitive confusion about disease symptoms, treatment and prognosis, and quantifies the degree of uncertainty through 25 items. Before the assessment, the researchers explained the contents of the items in detail to help the patients complete the score based on their real feelings, so as to reduce the understanding bias. To ensure data quality, research and implement multi-level quality control measures. All questionnaires and scales were filled out to avoid interference periods such as morning rounds, rehabilitation training and intensive treatment, so as to minimize the influence of external factors on patients'responses.

To clearly define sleep quality, according to the Richards Campbell Sleep Questionnaire (RCSQ) score, sleep quality is divided into three levels: good (≥ 70 mm), poor (50–69 mm), and poor (≤ 49 mm) [14]. This classification standard is based on previous research and the specific situation of the sample in this study. In addition, it is worth noting that RCSQ is mainly used to evaluate patients'subjective perception of sleep quality, and is not directly used for diagnosing sleep disorders. When conducting regression analysis, the total score is used as the main indicator to measure sleep quality. This method can more accurately capture the degree to which different factors affect sleep quality.

Data were collected through a standardized process from three sources:

Patient-provided data: Participants completed three self-reported surveys during 07:00–09:00 h: the Richards-Campbell Sleep Questionnaire (RCSQ) assessing previous night's sleep quality dimensions, the Perceived Social Support Scale, and the Chinese version of the Mishel Uncertainty in Illness Scale. Trained researchers provided standardized instructions and non-guided assistance.

Electronic Health Record (EHR) data: Researchers extracted demographic variables (gender, age, education, occupation, marital status, income), health data (BMI, primary diagnosis, comorbidities, payment method, smoking/alcohol history), and clinical data (APACHE II score, Charlson Comorbidity Index, mechanical ventilation status, indwelling tubes, sedative/analgesic administration history) from the hospital EHR system.

Researcher-collected data: Researchers performed the following:

Continuous environmental monitoring from 22:00–06:00 using TES-1351B sound level meters (30–130 dB range, ± 0.1 dB resolution) and TES-1335 digital illuminance meters (0–2000 Lux range, ± 3% accuracy) positioned 30 cm above the patient's head, differentiating between night-light and overhead lighting.

Recording the frequency of nocturnal nursing/medical interventions. For noise monitoring, hourly assessments of perceived noise annoyance were conducted using a validated annoyance scale, alongside A-weighted equivalent continuous sound pressure levels (Leq) to quantify average exposure. For light, mean illuminance (Lux) and duration of exposure > 100 Lux (indicating clinical interventions) were computed. Data processing excluded transient artifacts (e.g., calibration checks) using manufacturer-defined thresholds.

Recording physical restraint use (yes/no).

Assessing pain intensity using the Numerical Rating Scale (NRS).

Assessing consciousness using the Glasgow Coma Scale (GCS) and delirium using the CAM-ICU.

Quality control: Rigorous methodological safeguards were implemented across all study phases to ensure data integrity. All research personnel underwent standardized protocol training to ensure consistent execution of measurements and assessments. Instrument validity was verified through daily pre-use calibration of monitoring devices using certified reference standards. Structured operational procedures governed the timing of data collection to minimize confounding by clinical activities. Clinically assessed parameters adhered to validated scoring criteria administered by trained personnel. Survey instruments were administered following scripted instructions to minimize interviewer bias, with immediate verification for completeness and logical coherence of responses; ambiguous entries were rectified through direct participant clarification. Dual independent documentation and cross-verification of core outcome metrics were mandated, with any discrepancies exceeding predetermined thresholds triggering source data audit. Systematic auditing of data capture processes and compliance with ethical guidelines completed the framework. This multi-faceted approach enforced measurement precision, procedural uniformity, and analytical reliability throughout the investigation. The complete data collection and quality control process are shown in Fig. 2.

Data collection and quality control process. Include explanations of the abbreviations in the figure legends

SPSS version 27.0 was used for data analysis. Description statistics such as mean ± standard deviation for normally distributed data and median and quartiles fornon-normally distributed data were used to summarise the data. Count data were expressed as frequency, and percentage. In the univariate analysis exploring factors affecting sleep quality, the independent variables were gender and educational level. To identify potential variables related to sleep quality, appropriate statistical methods are selected based on variable types in univariate analysis. For binary variables such as gender, independent sample t-test is used to compare the differences between groups; For multi categorical variables such as education level, use one-way analysis of variance (ANOVA); If the data does not follow a normal distribution, Kruskal Wallis H test is used for non parametric testing. For continuous variables such as age and BMI, univariate linear regression analysis is used to evaluate their linear relationship with sleep quality. All statistical tests were conducted using P < 0.100 as the criterion for inclusion in the multiple regression model. Given that multiple comparisons may increase the risk of Type I errors, this study primarily focuses on pre-set assumptions and controls for confounding factors in a multiple regression model; During the research process, the researchers supervised the questionnaire filling on site to ensure complete data collection. Data analysis only uses complete data and does not employ interpolation methods.

The main outcome variable of this study is the subjective sleep quality of adult ICU patients, measured by the Richards Campbell Sleep Questionnaire (RCSQ) total score and its five sub dimensions (sleep depth, time to fall asleep, nighttime awakening, falling asleep again, and overall sleep quality) scores. Predictive variables include pain intensity, disease uncertainty, number of nighttime treatment procedures, use of physical constraints, and duration of nighttime light exposure. Based on previous research indicating that these factors may have an impact on sleep quality, this study hypothesized a significant correlation between the predicted variables and patient sleep quality, and validated this hypothesis through a single linear regression analysis.

The sample size is determined based on the guidelines for multiple linear regression analysis, and it is recommended to have 10 participants for each independent variable to test the equivalent stress (Cohen’s f2 = 0.15). This effect size was defined by Cohen in 1988 as the independent variable explaining approximately 13% of the variation in the dependent variable at 80% statistical power and alpha = 0.05 level [15]. Specifically, it refers to the magnitude of the effect between the independent variables included in this study (including pain level, disease uncertainty, frequency of nighttime diagnosis and treatment operations, use of restraints, duration of nighttime light exposure, etc.) and the dependent variables (scores and total scores of various dimensions such as sleep depth, sleep onset latency, nighttime awakening, falling asleep again, and overall sleep quality measured by the RCSQ scale). Where f2 is equal to the ratio of explained variance (R2) to unexplained variance (1-R2). When f2 = 0.15, it indicates that the independent variables can jointly explain about 13% of the variation of the dependent variable (R2 ≈ 0.13). Consider 18 independent variables with a minimum sample size of 180 cases. Taking into account a 10% non response rate, the target sample size is adjusted to 198 cases. In addition, post hoc verification calculations confirmed that the final 172 included samples had a testing power of 0.82, sufficient to detect the predetermined moderate effect size (f2 = 0.15).

This study has been reviewed and approved by the Ethics Committee of The First Affiliated Hospital of Nanjing Medical University (2023-SR-050). Participants were fully informed that their involvement in the study was completely voluntary, and they had the right to decline participation at any time without providing a specific reason. They were also informed about the study’s purpose, procedures, potential risks, and benefits before providing signed informed consent.