Association of surgical timing with 30-day independent ambulation and postoperative sleep disorder in older adult patients undergoing joint replacement: an exploratory secondary analysis of a prospective cohort

Abstract

Objective

This exploratory study investigated the observational association between surgical timing (morning versus afternoon) and 30-day independent ambulation in older adult joint replacement patients, and whether postoperative sleep disorder (PSD) mediates this association.

Methods

This was an exploratory secondary analysis of a prospectively registered cohort. Patients aged ≥ 65 years undergoing joint replacement were included. Surgical timing was the exposure, 30-day independent ambulation the primary outcome, and PSD the potential mediator. Propensity score matching (PSM) was used to balance baseline characteristics, and a doubly robust sensitivity analysis was performed to address residual imbalance. Logistic regression and counterfactual mediation analysis were then conducted.

Results

Among 753 eligible patients, PSM balanced most baseline characteristics. Afternoon surgery was significantly associated with higher odds of 30-day independent ambulation (OR = 2.106, 95% CI: 1.407–3.146, p = 0.001). The Natural Direct Effect remained significant (OR = 2.174, p < 0.001), while the Natural Indirect Effect via PSD was not significant (OR = 0.969, p = 0.654). E-value analysis suggested moderate robustness.

Conclusion

Afternoon surgery is associated with better 30-day independent ambulation, and PSD does not mediate this observational association. All findings are exploratory, hypothesis-generating, and require validation in randomized controlled trials.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12877-026-08015-3.

Introduction

The global population is aging at an unprecedented rate, with projections indicating that by 2050, adults aged 65 years and older will account for over 20% of the global population [1]. This demographic shift has been accompanied by a marked rise in age-related musculoskeletal conditions, particularly degenerative joint diseases such as osteoarthritis [2–4]. For older adult patients with severe joint impairment, joint replacement surgery has emerged as a cornerstone of treatment, effectively alleviating pain, restoring mobility, and enhancing overall quality of life. However, despite its proven efficacy, postoperative outcomes in this population remain variable, with functional recovery, most critically the ability to walk independently, and the incidence of complications serving as key indicators of surgical success [5–7].

Among the complications commonly observed after joint replacement surgery, postoperative sleep disorder (PSD) stands out as a prevalent and underrecognized issue [8–10]. Studies have reported that older adult surgical patients experience sleep disruptions in the immediate postoperative period, characterized by difficulties falling asleep, frequent awakenings, and reduced sleep duration or quality [11, 12]. Beyond the immediate discomfort, such disturbances have far-reaching implications: impaired sleep has been linked to delayed wound healing, heightened pain perception, increased risk of delirium, and compromised immune function, all of which can hinder the recovery of motor function, including the ability to walk independently [13–17]. For older adult patients, whose physiological reserves are already diminished, these effects may be amplified, making sleep quality a critical, yet often overlooked, determinant of postoperative recovery.

Against this backdrop, the timing of surgical intervention has garnered increasing attention as a potential modulator of postoperative outcomes. Circadian rhythms, the body’s internal 24-hour cycles regulating physiological processes such as hormone secretion, pain sensitivity, and immune function, play a pivotal role in how the body responds to stress, including surgical trauma. Morning and afternoon surgeries may thus differ in their impact on patient recovery: for instance, cortisol levels, which peak in the morning, could influence stress responses and inflammation, while afternoon procedures might align better with the body’s natural rhythms of pain tolerance and tissue repair. However, existing research on surgical timing in joint replacement has primarily focused on short-term outcomes such as operative duration, blood loss, and infection rates, with limited exploration of its effects on patient-reported outcomes like sleep quality or functional recovery, particularly in older adult populations [18–21].

This gap is notable given the unique vulnerability of older adults to both surgical stress and sleep disturbances. Age-related changes in circadian regulation, coupled with comorbidities and polypharmacy, make older adult patients more susceptible to sleep disruptions postoperatively. Yet, whether surgical timing exacerbates or mitigates these disruptions, and how such effects, in turn, influence the recovery of independent ambulation, remains poorly understood. Clarifying this relationship could provide valuable insights into optimizing perioperative care for this growing patient population.

To address this knowledge gap, the present study aims to investigate the observational association between surgical timing and the ability of older adult patients to walk independently one month after joint replacement surgery. As an exploratory secondary analysis of a prospectively registered parent cohort initially focused on PSD incidence, this study additionally seeks to examine whether PSD mediates the relationship between surgical timing and functional recovery. By rigorously adjusting for baseline imbalances, this research endeavors to offer hypothesis-generating insights into refining surgical scheduling and improving functional outcomes for older adult joint replacement patients.

Methods

Study design

This study is an exploratory secondary analysis of a prospectively registered cohort (Chinese Clinical Trial Registry: ChiCTR2300074920). The study protocol was approved by the Ethics Review Committee of the First Affiliated Hospital of Guangxi Medical University (No. 2023-K182-01). Informed consent was obtained from all participants or their legal guardians. All study procedures adhered to the Declaration of Helsinki.

Transparency Declaration Regarding Registry Discrepancies: We transparently acknowledge distinct deviations between the initial clinical registry and the present manuscript. First, regarding the study timeframe, data collection commenced in May 2023 following ethical approval; however, due to administrative oversight, formal public registration on the ChiCTR platform was delayed until August 2023. Second, regarding the cohort size, the originally registered sample size of 500 represented the minimum requirement. To enhance statistical power for this secondary mediation framework, we clinically elected to extend enrollment to 1216 screened cases, with 753 ultimately included post-quality control. Third, while the parent protocol pre-specified postoperative sleep disorder as the primary focus, the evaluation of surgical timing as an exposure and 30-day independent ambulation as the primary functional outcome were formulated post hoc as exploratory analyses to address emerging clinical questions. Consequently, this study makes no strict causal claims and is strictly framed as an exploratory observational association study.

Finally, regarding the funding source discrepancy, while the initial registry entry listed the funding as ‘self-raised’ during the early inception of the study, the project subsequently received formal grant support as detailed in the Funding disclosure of this manuscript.

Study participants

The study included patients aged 65 years or older who were scheduled to undergo elective total hip arthroplasty or total knee arthroplasty under general anesthesia. The inclusion criteria were as follows: aged 65 years or older; American Society of Anesthesiologists physical status classification ranging from I to IV; and scheduled for elective surgery under general anesthesia. The exclusion criteria were as follows: preoperative history of dementia or psychiatric disorders; and incomplete medical records. A total of 1216 consecutive eligible patients were screened during the study period; 753 were finally included for data analysis after excluding cases who denied consent, were unable to communicate, had Parkinson’s or Alzheimer’s disease, had their surgery canceled, were admitted to the ICU postoperatively, died during the first postoperative days, or were undergoing secondary or revision TJA, as detailed in Fig. 1.

Fig. 1.

Flowchart of initial patient selection for the study. The flowchart details the initial screening and exclusions resulting in the pre-matching cohort of 753 patients. A subsequent 1:1 Propensity Score Matching (PSM) was applied to this cohort to balance baseline covariates, resulting in the final matched cohort of 552 patients (detailed in Table 1 and Supplementary Figure S1) used for the core counterfactual mediation analysis

Data collection and variable definition

Preoperative demographic and clinical characteristics

Data on patients’ age, gender, body mass index, educational level, and American Society of Anesthesiologists physical status classification were extracted from the hospital’s electronic medical record system, which included inpatient medical records and medical history collection records. Preexisting comorbidities such as hypertension, diabetes mellitus, and cardiovascular diseases were confirmed based on the “past medical history” records in the medical charts. Smoking history and drinking history were extracted from the “social history” or “personal history” sections. Data related to preoperative sleep disturbance were obtained from preoperative evaluations, and the Sleep Numerical Rating Scale and Athens Insomnia Scale were used to determine whether patients had baseline preoperative sleep disturbance. Patients’ frailty status was assessed using the FRAIL scale, which evaluates fatigue, resistance, ambulation, illness, and weight loss. Based on the FRAIL scale scores, patients were divided into three categories: robust with a score of 0, pre-frail with a score of 1 to 2, and frail with a score greater than 2. Preoperative pain intensity was recorded using the Visual Analogue Scale, including resting pain score (VAS0) and activity pain score (VAS1). Anxiety symptoms were assessed using the 7-item Generalized Anxiety Disorder Scale, and a score greater than 4 was defined as the presence of anxiety. Metabolic syndrome was determined based on laboratory test results and physical examination findings in the medical records. It was defined as meeting at least three of the following criteria: obesity, elevated fasting triglycerides, reduced high-density lipoprotein cholesterol, elevated blood pressure, and elevated blood glucose

Definition of PSD

PSD was defined based on the combined assessment of the Sleep Numerical Rating Scale (sleep-NRS) and Athens Insomnia Scale (AIS) at 1 day, 3 days, and 7 days postoperatively. The diagnostic criteria were as follows: Sleep-NRS: A score ≥ 4 (0 = no sleep disturbance, 10 = worst possible sleep disturbance) indicated clinically significant sleep disturbance; AIS: A total score ≥ 6 indicated insomnia; PSD was confirmed if either the sleep-NRS score met the threshold OR the AIS score met the threshold at any one of the three measurement time points (postoperative day 1, 3, or 7).

This definition may introduce misclassification bias: it may inflate the incidence of PSD and does not reflect symptom severity or persistence. Combining two different instruments without clear harmonization is also a methodological limitation. This pragmatic approach was chosen for clinical feasibility in the original cohort design, and its limitations are explicitly acknowledged. Sensitivity analysis using a stricter PSD definition (positive at both POD 1 and POD 3) yielded consistent results with the primary analysis.

Preoperative laboratory test data

All data are extracted from the laboratory test reports in the medical records and are the test results within 72 h before surgery. If a patient has multiple complete blood routine reports, the one closest to the surgery is selected. To reduce the interference of irrelevant confounders on the analytical model, we mainly focus on the Systemic Immune-Inflammation Index (SII), with the calculation formula: SII = platelet count × neutrophil count / lymphocyte count; and the Prognostic Nutritional Index (PNI), with the calculation formula: PNI = serum albumin (g/L) + 5 × lymphocyte count (×10⁹/L).

Perioperative management

All patients received standardized general anesthesia and multimodal perioperative analgesia according to our institutional protocol. General anesthesia was induced and maintained using standard weight-based dosing. Postoperative analgesia included patient-controlled intravenous analgesia (PCIA) and scheduled non-steroidal anti-inflammatory drugs (NSAIDs) to maintain resting VAS scores below 4. Standardized postoperative rehabilitation, including bedside sitting and early mobilization, was uniformly encouraged for all patients guided by the surgical team.

Intraoperative data

Intraoperative data were extracted from anesthesia records, surgical nursing records, and surgical inventory records. Specific data included the start time of surgery, which was divided according to the time recorded in the medical records—morning from 08:00 to 12:00 and afternoon from 14:00 to 18:00. Cases where surgery commenced between 12:00 and 13:59 or outside the predefined blocks were excluded to maintain two distinct, non-overlapping physiological blocks and avoid the transitional midday circadian phase. Other specific data included duration of surgery; intraoperative blood loss; intraoperative fluid infusion volume; intraoperative blood transfusion status; use of dexmedetomidine; occurrence of intraoperative hypotension, which was defined as a mean arterial pressure decrease of more than 20% compared with the preoperative baseline value or a mean arterial pressure lower than 60 mmHg; intraoperative analgesic dosage; implementation of peripheral nerve block; and intraoperative preparation of patient-controlled analgesia devices. Patients were divided into the morning group (Group M) and the afternoon group (Group A) according to the start time of their surgery.

Outcome measure definition

The primary outcome measure was the patient’s independent ambulation ability at one month after surgery, assessed via telephone follow-up on the 30th postoperative day. Independent ambulation was operationally defined as the ability to walk continuously for ≥ 10 m without human assistance or reliance on assistive devices. For patients unable to be interviewed directly, designated family informants provided the information.

This outcome is a pragmatic, real-world functional milestone rather than a finely graded validated clinical scale. This is a major methodological limitation: precise preoperative walking distance and baseline functional status were not available for adjustment, and baseline functional status is one of the strongest predictors of postoperative mobility. Frailty assessment alone cannot fully account for baseline functional differences. Achieving unsupported ambulation at 30 days remains a meaningful proxy for early functional independence in the older adult arthroplasty population, but the lack of validated baseline functional adjustment must be acknowledged as a critical constraint.

Data preprocessing

For the initial clinical features (including scale scores at various time points and laboratory test indicators) of 753 patients, all data were complete with no missing values for any variables; therefore, no missing value imputation was performed. We first performed feature integration and data transformation: laboratory test indicators were merged into the Systemic Inflammatory Index (SII) and Prognostic Nutritional Index (PNI) for comprehensive evaluation, while specific scale scores were converted into categorical data. After these processes, 23 clinical features were actually included in the subsequent mediation analysis.

Covariates were selected a priori based on clinical relevance and established literature regarding perioperative sleep and functional recovery. These pre-exposure confounders included age, gender, body mass index, educational level, ASA physical status classification, procedure type (THA or TKA), baseline comorbidities (e.g., hypertension, diabetes, coronary heart disease, metabolic syndrome), and preoperative sleep status. Crucially, purely data-driven covariate selection methods were avoided, and post-exposure variables were strictly excluded to prevent collider bias. Key perioperative variables including anesthesia technique, multimodal analgesia, rehabilitation intensity, surgical complexity, and OR workflow were unavailable and not adjusted.

Data analysis

Baseline data analysis

Normality tests were performed on continuous variables. Continuous variables were presented as mean ± standard deviation or median with interquartile range, depending on their distribution. Student’s t-test or Mann–Whitney U test was used for comparison between groups. Categorical variables were reported as counts with percentages and were analyzed using the chi-square test or Fisher’s exact test. Statistical significance was defined as a two-tailed P-value less than 0.05.

Propensity score matching and mediation analysis

Given the observational nature of surgical scheduling, substantial baseline imbalances existed between the morning and afternoon groups, which severely threatened causal inference. To mitigate this selection bias, we implemented Propensity Score Matching (PSM). A logistic regression model was used to calculate propensity scores based on all available pre-exposure covariates (Age, BMI, Gender, Education, ASA status, baseline comorbidities, Preoperative sleep disorder, Pre_Anxiety, Frail status, Procedure Type, Systemic Immune-Inflammation Index [SII], and Prognostic Nutritional Index [PNI]). Patients were matched 1:1 using a nearest-neighbor greedy matching algorithm with a strict caliper of 0.05, without replacement. Following PSM, mediation analysis was conceptualized using a modern counterfactual causal mediation framework on the highly balanced matched cohort. Effects were estimated on the odds ratio (OR) scale using a regression-based approach with 1000 bootstrap resamples for 95% confidence interval (CI) estimation. The total effect was decomposed into a Natural Direct Effect (NDE) and a Natural Indirect Effect (NIE). Crucially, post-exposure variables were strictly excluded from the models to prevent collider bias. Although PSM improved balance significantly, slight residual imbalance remained in PNI, age, and ASA status; we performed additional adjustment after matching using a doubly robust approach in sensitivity analyses to further mitigate residual confounding.

Sensitivity analysis

Two complementary sensitivity analyses were conducted. First, causal E-values were calculated for the observed ORs (Total Effect, NDE, and NIE) using the standard formula by VanderWeele & Ding (2017): E-value = OR + √[OR × (OR − 1)] for effects > 1. The E-value quantifies the minimum strength of association on the risk ratio scale that an unmeasured confounder would need to have with both the exposure and the outcome to fully explain away the specific effect. Second, to address the robustness of outcome and mediator definitions, sensitivity models were performed by (1) redefining PSD strictly as meeting thresholds at both day 1 and day 3 postoperatively, and (2) broadening the independent ambulation outcome to allow the use of a single-point cane. Sensitivity analysis using the stricter PSD definition confirmed no significant mediation effect (data not shown), supporting the stability of the primary mediation findings.

Mediation analysis assumptions

The counterfactual mediation analysis relies on two key untestable assumptions: (1) no unmeasured confounding between the mediator (PSD) and outcome (30-day independent ambulation); (2) no exposure-induced mediator–outcome confounders. These key assumptions are unlikely to be fully satisfied in this observational dataset due to unmeasured perioperative confounders. Therefore, all mediation results are strictly exploratory and hypothesis-generating, with no confirmatory causal implications.

Results

General information and propensity score matching

A total of 753 older adult patients who underwent joint replacement surgery were initially evaluated in this study, among which 362 (48.1%) were in the morning group and 391 (51.9%) were in the afternoon group. To assess potential selection bias from attrition, baseline characteristics were compared between the included cohort (n = 753) and patients excluded due to the clinical or administrative reasons detailed in Fig. 1 (n = 463). No significant differences were observed in key demographics, suggesting minimal selection bias (Supplementary Table S1).

Given the observational nature of surgical scheduling, significant baseline imbalances existed between the morning and afternoon groups in the crude cohort. To address this, 1:1 PSM was successfully executed. After matching, a highly balanced cohort of 552 patients was established, with 276 patients in the morning surgery group and 276 patients in the afternoon surgery group (Supplementary Figure S1). As shown in Table 1, the vast majority of baseline characteristics, including BMI, preoperative sleep disorder, frailty status, and comorbidity prevalence, were statistically balanced between the two groups. Although slight differences remained in pre-PNI, Age, and ASA status (Standardized Mean Difference > 0.1 in Supplementary Figure S1), the overall matching significantly improved baseline comparability, establishing a robust foundation for subsequent causal inference.

Table 1.

Baseline and Postoperative Characteristics of the Propensity Score-Matched Cohort

Variables	Total (N = 552)	Morning (n = 276)	Afternoon (n = 276)	Statistic	P-Value
Preoperative Baseline Characteristics
Age (years), Mean ± SD	71.25 ± 4.78	71.62 ± 5.11	70.89 ± 4.40	t = 1.79	0.074
BMI (kg/m²), Mean ± SD	24.52 ± 4.12	24.74 ± 4.13	24.30 ± 4.10	t = 1.24	0.215
Gender, n(%)				χ² = 0.08	0.784
Female	376 (68.12%)	186 (67.39%)	190 (68.84%)
Male	176 (31.88%)	90 (32.61%)	86 (31.16%)
Education level, n(%)				χ² = 1.72	0.422
Primary school or below	235 (42.57%)	115 (41.67%)	120 (43.48%)
Middle school	272 (49.28%)	142 (51.45%)	130 (47.10%)
High school or above	45 (8.15%)	19 (6.88%)	26 (9.42%)
ASA physical status, n(%)				χ² = 1.72	0.190
II	338 (61.23%)	161 (58.33%)	177 (64.13%)
III	214 (38.77%)	115 (41.67%)	99 (35.87%)
Frail status, n(%)				χ² = 0.90	0.639
Robust	124 (22.46%)	63 (22.83%)	61 (22.10%)
Pre-frail	310 (56.16%)	150 (54.35%)	160 (57.97%)
Frail	118 (21.38%)	63 (22.83%)	55 (19.93%)
Procedure Type, n(%)				χ² = 0.47	0.494
THA	253 (45.83%)	122 (44.20%)	131 (47.46%)
TKA	299 (54.17%)	154 (55.80%)	145 (52.54%)
Comorbidities & Risk Factors, n(%)
History of Smoking (Yes)	59 (10.69%)	29 (10.51%)	30 (10.87%)	χ² = 0.00	1.000
History of Drinking (Yes)	72 (13.04%)	35 (12.68%)	37 (13.41%)	χ² = 0.02	0.899
Hypertension (Yes)	275 (49.82%)	138 (50.00%)	137 (49.64%)	χ² = 0.00	1.000
Diabetes (Yes)	62 (11.23%)	28 (10.14%)	34 (12.32%)	χ² = 0.45	0.500
Coronary heart disease (Yes)	34 (6.16%)	18 (6.52%)	16 (5.80%)	χ² = 0.03	0.859
Cerebrovascular disease (Yes)	41 (7.43%)	20 (7.25%)	21 (7.61%)	χ² = 0.00	1.000
Metabolic syndrome (Yes)	201 (36.41%)	101 (36.59%)	100 (36.23%)	χ² = 0.00	1.000
Preoperative sleep disorder (Yes)	200 (36.23%)	101 (36.59%)	99 (35.87%)	χ² = 0.01	0.929
Preoperative anxiety (Yes)	116 (21.01%)	60 (21.74%)	56 (20.29%)	χ² = 0.10	0.754
Preoperative Laboratory Data
SII, Mean ± SD	691.33 ± 370.09	697.93 ± 352.49	684.74 ± 387.41	t = 0.42	0.676
PNI, Mean ± SD	48.68 ± 4.44	49.14 ± 4.41	48.21 ± 4.43	t = 2.47	0.014
Postoperative Outcomes
Ambulation Day, Median (Q1, Q3)	2 (2, 3)	2 (2, 3)	2 (1, 3)	Z = -2.61	0.009
PSD Incidence (Yes), n(%)	250 (45.29%)	139 (50.36%)	111 (40.22%)	χ² = 5.33	0.021
Independent ambulation (Yes), n(%)	344 (62.32%)	148 (53.62%)	196 (71.01%)	χ² = 17.04	< 0.001

Postoperative outcomes in the matched cohort

Regarding the primary outcome within the matched cohort, a significant difference was observed between the two groups. The afternoon surgery group demonstrated a significantly higher rate of independent ambulation at 30 days compared to the morning group (71.01% vs. 53.62%, χ²=17.04, p < 0.001). Baseline functional walking status was not available for adjustment in this analysis, which represents a major methodological constraint. Regarding the primary mediator, the overall incidence of postoperative sleep disorder (PSD) was significantly lower in the afternoon group (40.22%) compared to the morning group (50.36%, χ²=5.33, p = 0.021). Furthermore, the initiation of postoperative ambulation was significantly earlier in the afternoon group (Median 2 days [Q1-Q3: 1–3] vs. 2 days [Q1-Q3: 2–3], Z=-2.61, p = 0.009).

Mediation analysis on matched cohort

To rigorously explore whether PSD mediates the association between surgical timing and functional recovery, a counterfactual mediation framework was applied to the matched cohort (Table 2; Fig. 2). The analysis revealed a highly significant total association: afternoon surgery was associated with higher odds of 30-day independent ambulation (OR = 2.106, 95% CI: 1.407–3.146, p = 0.001).

Table 2.

Counterfactual Mediation and Sensitivity Analysis of Surgical Timing on 30-Day Independent Ambulation

Effect Path	Estimated OR (95% CI)	P-Value	E-value (Estimate)	E-value (Lower CI Limit)
Total Effect (c)	2.106 (1.407, 3.146)	0.001	3.63	2.16
(Surgical Timing → Independent ambulation)
Natural Direct Effect (NDE)	2.174 (1.509, 3.302)	0.001	3.77	2.38
(Surgical Timing → Walking, independent of PSD)
Natural Indirect Effect (NIE)	0.969 (0.782, 1.128)	0.654	1.21	1.00
(Surgical Timing → PSD → Walking)

The E-value indicates the minimum strength of association that an unmeasured confounder would need to have with both the exposure and the outcome to fully explain away the observed association. For non-significant results where the 95% confidence interval includes the null value (OR = 1), the E-value for the confidence interval limit is defined as 1.00, meaning no unmeasured confounding is needed to render the confidence interval inclusive of the null

Fig. 2.

Conceptual Directed Acyclic Graph (DAG) for the Proposed Causal Structure. The diagram illustrates the theoretically assumed pathways between surgical timing (Exposure), postoperative sleep disorder (Mediator), and 30-day independent ambulation (Outcome). The dashed lines indicate the baseline confounders that were adjusted via Propensity Score Matching (PSM) to estimate the Natural Direct Effect (NDE) and Natural Indirect Effect (NIE) under the counterfactual framework

When decomposing this association, the Natural Direct Effect (NDE) remained strong and significant (OR = 2.174, 95% CI: 1.509–3.302, p = 0.001), indicating that afternoon surgery was associated with improved ambulation independently of postoperative sleep quality. The Natural Indirect Effect (NIE) via PSD was not statistically significant after PSM adjustment (OR = 0.969, 95% CI: 0.782–1.128, p = 0.654). This indicates that when baseline demographic and clinical differences are rigorously balanced, PSD does not serve as a significant mediating pathway between surgical timing and early functional recovery.

Sensitivity analysis

In our causal E-value sensitivity analysis (Table 2), the observed Total Effect yielded an E-value of 3.63 (lower CI limit = 2.16). The Natural Direct Effect yielded an E-value of 3.77 (lower CI limit = 2.38), indicating moderate robustness of the direct association. However, these E-values do not rule out confounding from clinically plausible unmeasured variables, and causal inference remains limited. While the indirect mediation pathway was not statistically significant (point estimate E-value = 1.21, lower CI limit E-value = 1.00 as the interval includes the null), the E-value for the NDE suggests moderate robustness but does not rule out confounding from clinically plausible unmeasured variables. Additionally, doubly robust analysis (regression adjustment after PSM) was conducted to address residual imbalance in PNI, age, and ASA status. This analysis yielded consistent results (OR = 2.152, 95% CI: 1.431–3.204, p < 0.001), further supporting the primary observational association.

Discussion

This exploratory secondary analysis examined the observational association between surgical timing, PSD, and 30-day independent ambulation in older adult joint replacement patients. Crude analyses suggested a potential mediation pathway involving PSD. However, upon rigorous PSM to correct for baseline clinical imbalances, this mediation association was not statistically significant. Instead, we identified a robust direct observational association: afternoon surgery was associated with higher odds of independent ambulation, independent of postoperative sleep quality.

The lack of significant PSD mediation underscores the vulnerability of observational mediation models to baseline confounding. Crude associations between sleep and walking recovery are likely driven by preoperative patient characteristics rather than a direct causal mechanism triggered by surgical timing [22, 23].

The strong Natural Direct Effect of afternoon surgery on functional recovery remains a compelling observational finding [24–27]. This non-sleep-dependent association may be hypothetically rooted in human circadian biology (e.g., cortisol levels, pain tolerance, cardiovascular stability). These are purely hypothetical mechanisms and are NOT directly supported by the present study data. These hypotheses are interesting but are not directly supported by the present data and should not be interpreted as mechanistic conclusions [28, 29]. Afternoon surgeries might also align with institutional perioperative workflows, such as staffing patterns, physical therapy availability, or anesthesia protocols, which could create a more favorable environment for early motor function recovery [30–33].

In our revised analysis, post-exposure variables such as the first postoperative ambulation day were strictly excluded from the primary mediation models to prevent collider bias and ensure the validity of causal estimates [34–37]. Although early ambulation is traditionally recognized as a key factor in functional recovery, its inclusion as a confounder in a causal mediation framework would be methodologically inappropriate. Regarding pre-exposure confounders, the ASA classification showed a significant association with walking ability but no correlation with PSD [38–42]. Other adjusted baseline clinical factors showed no statistically significant impact in the final mediation models.

The clinical significance of this study lies in providing observational insights into the potential advantages of afternoon surgical timing in two dimensions: reducing the risk of PSD and promoting postoperative walking. It provides a preliminary reference for surgical scheduling. While clinicians may consider afternoon surgery based on individual patient conditions, these findings should be validated by randomized trials. Additionally, the potential role of Troponin T as a predictor of poor prognosis in older adult functional recovery warrants specialized investigation in future studies [43].

However, this study has several critical limitations. First, residual and unmeasured confounding remain unavoidable despite PSM. Key unmeasured perioperative factors—including anesthesia technique, multimodal analgesia, rehabilitation intensity, surgical complexity, and operating room workflow differences—may bias results and limit causal inference. Second, the primary outcome is a pragmatic milestone without adjustment for baseline functional status, a major predictor of postoperative mobility, and frailty cannot substitute for baseline walking ability. Third, PSD definition may introduce misclassification bias, which may inflate incidence and fail to reflect symptom severity or persistence. Fourth, counterfactual mediation analysis relies on two untestable key assumptions that cannot be verified in this observational dataset. Fifth, residual imbalance remained in some variables after PSM, and E-values only indicate moderate robustness without ruling out clinically plausible confounding. All conclusions are strictly observational and hypothesis-generating.

Conclusion

In this exploratory secondary analysis, afternoon surgical timing was observationally associated with higher odds of 30-day independent ambulation in older adult joint replacement patients. Postoperative sleep disorder did not significantly mediate this observational association. Due to the observational design, residual confounding, and unmeasured perioperative factors, all findings are strictly hypothesis-generating and require prospective randomized controlled trials for definitive confirmation.

Authors’ contributions

LLH: Conceptualization, Methodology, Investigation, Data Collection, Data Curation, Writing – Original Draft, Writing – Review & Editing. GH: Conceptualization, Methodology, Data Collection, Writing – Original Draft. CLN and WY: Investigation, Methodology, Writing – Original Draft. WY: Conceptualization, Supervision, Review. XYB: Conceptualization, Supervision, Data Curation, Writing – Review & Editing and Obtain funding.

Funding

This work was supported by Guangxi Science and Technology Base and Talent Special Project (No. AD25069060) and Guangxi Key Research and Development Program (No. AB24010066).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

The study was performed in strict accordance with the Declaration of Helsinki. The study protocol was approved by the Medical Ethics Committee of the First Affiliated Hospital of Guangxi Medical University, China (Identifier: NO.2023-K182-01). The informed consents to participate in the study had been obtained from all participants or their legal guardian(s). Where all data were completely de-identified. The study was registered at the Chinese Clinical Trial Registry (www.chictr.org.cn, Registration number: ChiCTR2300074920, at 2023-08-21).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.