The Confusion Assessment Method for the Intensive Care Unit (CAM-ICU) (17) and the CAM (18) were used to assess delirium once daily for post-operative 5 days by nurses in the ICU and ward. Both CAM and CAM-ICU contain 4 criteria: acute onset of fluctuating course, inattention, disorganized thinking, and altered consciousness, while satisfying criteria 1 and 2 and either 3 or 4, it was positive. Meanwhile, HH and JG reviewed the medical records of each patient in detail, found patients whose symptom description matched the Diagnostic and Statistical Manual of Mental Disorders-5 (DSM-5) (3) through a chart review, and confirmed the diagnosis when they had the same judgments. In patients older than 65 years, the anesthesiologist in the institution routinely assessed cognitive function using the Mini-Mental State Examination (MMSE) (19) during pre-operative visits; if the patient had the MMSE score of <27, it was recorded as having mild cognitive impairment. Finally, patients without mild cognitive impairment were categorized into the POD and non-POD groups.

Laboratory Hb concentrations were monitored at admission and on post-operative day 1 according to the diagnostic procedure and, if necessary, according to the clinical judgment about re-examination. In this study, we extracted Hb concentrations at admission and on post-operative day 1. The change in Hb concentration was the absolute value of the difference in pre-operative and post-operative Hb concentrations.

Demographic characteristics including sex, age, body mass index, smoking and drinking history, and the American Society of Anesthesiology (ASA) score were collected. Comorbidity burden included major heart disease, lung disease, metabolic disease, and cerebrovascular disease. Hypoalbuminemia was defined as an albumin level of 35 g/l. The units of RBC transfusion from admission to post-operative 5 days were collected. Data on the operation and anesthesia time, intraoperative blood loss, types of operation and anesthesia, and post-operative ICU transfer were collected. Primary acute post-operative complications such as hypoxemia, electrolyte imbalance, and arrhythmia were collected. In addition, information on perioperative drug use and LOS was collected.

Patients were encouraged to stand and walk gradually from post-operative day 1 or mechanically-assisted rehabilitation was performed in bed. Post-operative pain assessment using the visual analog scale (VAS) was conducted by a trained nurse once daily within post-operative 3 days. The pain score was divided into four grades according to pain intensity: painless (0 point), mild (1–3 points), moderate (4–6 points), and severe (7–10 points) pain.

Regarding statistical analysis, categorical variables are expressed as frequencies and percentages, normally distributed continuous variables are expressed as mean and SD, and non-normally distributed continuous variables are expressed as median and interquartile range. With respect to continuous variables, Student's t-test was performed to test for differences between groups. The chi-squared or the Wilcoxon signed-rank tests were performed to detect differences between groups for discontinuous variables or non-normally distributed continuous variables. Patients with missing records of Hb concentrations, blood transfusion, delirium, or cognitive assessment were excluded.

Age, stroke history, pre-operative and post-operative Hb concentrations, perioperative changes in Hb concentrations, intraoperative blood loss, surgery and anesthesia type, surgery and anesthesia time, pre-operative and post-operative hypoalbuminemia, post-operative ICU transfer, post-operative hypoxemia, and the VAS scores within post-operative 3 days were analyzed using the multivariate logistic regression model. Sensitivity analyses were performed using the multivariate logistic regression with the variables deleted. The product term of pre-operative Hb concentration and change in Hb concentration [Hb (pre) × Hb change] reflects the interaction in the logistic regression analysis. Therefore, we conducted 4 models: model 1: logistic regression analysis contained Hb (pre- and post-operative), perioperative Hb change, age, stroke history, surgery and anesthesia type, hypoalbuminemia (pre- and post-operative), midazolam, intraoperative blood loss, operation time, anesthesia time, ICU transfer, hypoxemia, and VAS in 1–3 days after surgery; model 2: logistic regression analysis contained factors in model 1, except perioperative Hb change; model 3: logistic regression analysis contained factors in model 1, except pre-operative Hb concentration; model 4: logistic regression analysis contained factors in model 1 and factor “Hb (pre) × Hb change.” Statistical significance was set at a two-tailed p-value of 0.05. Data were coded and stored and analyzed using the Statistical Package for the Social Sciences (SPSS) version 22.0 (SPSS Incorporation, Chicago, Illinois, USA).

A total of 172 patients (19.3%) developed POD. Of these patients, 151 patients (87.8%) developed delirium within post-operative 3 days (Table 1). The mean age in the POD and non-POD groups was 81.3 ± 7.0 and 77.3 ± 7.3 years, respectively. A total of 646 women (72.7%) were finally included. In this study, 732 patients (82.3%) had the ASA 3 grade and no patient had the ASA 1 grade in this study. LOS was longer in the POD group than in the non-POD group (11.3 ± 5.1 vs. 10.5 ± 4.7 days, p = 0.078) (Table 2). No patient data was lost from the 889 participants. No patient was painless on the first day after surgery both in the POD and non-POD groups.

The pre-operative and post-operative Hb concentrations were lower in the POD group than in the non-POD group (116 ± 16.4 vs. 120.5 ± 16.6 g/l, p = 0.004; 96.2 ± 16.2 vs. 101.1 ± 15.9 g/l, p < 0.001). No patient had Hb concentration of <7 g/l pre-operatively, whereas 22 patients (2.5%) had post-operative Hb concentration of <7 g/l. The mean change in Hb concentration was 21.2 ± 12.3 g/l and the change in perioperative Hb concentration was higher in the POD group than in the non-POD group (23.1 ± 12.8 vs. 20.7 ± 12.2 g/l).

Low pre-operative Hb concentrations [p = 0.026, odds ratio (OR) = 0.978, 95% CI = 0.959–0.997] and change in perioperative Hb concentration (p = 0.006, OR = 1.033, 95% CI = 1.010–1.057) together were significantly associated with POD after adjustment for age, stroke history, surgery and anesthesia type, hypoalbuminemia (pre-operatively and post-operatively), midazolam use, intraoperative blood loss, operation and anesthesia time, post-operative ICU transfer and hypoxemia, and the VAS score within post-operative 3 days (Table 3). After excluding one of the two factors (pre-operative Hb concentration and change in Hb concentration), respectively, with other confounding factors left unchanged, another factor was no more significant (Table 4). Including the factor “Hb (pre) × Hb change” in the logistic regression analysis, there was no interaction effect between pre-operative Hb concentration and the change of Hb concentration. Moreover, the effect of the factor “Hb change” was significant in model 4 (p = 0.022, OR = 1.139, 95% CI = 1.019–1.274), while the effect of “Hb (pre)” was not.

The difference of the area under the curve (AUC) between models, which included pre-operative Hb concentration and Hb change, respectively, in the receiver operating characteristic (ROC) curve analysis, was fairly small (model 2: AUC = 0.745, p < 0.001, 95% CI = 0.703–0.787; model 3: AUC = 0.742, p < 0.001, 95% CI = 0.699–0.784) (Figure 2). Age (p < 0.001, OR = 1.072), stroke history (p = 0.003, OR = 2.489), post-operative ICU transfer (p = 0.007, OR = 1.981), and the VAS score within post-operative 2 days (p₁ = 0.016 and p₂ = 0.006) were independently associated with POD in the logistic regression analysis.

This study has some strengths. The sample size of 889 patients was large. As far as we know, this study is the first study to assess the relationship between change in Hb concentration and POD; This study found a positive association between the two.

This study has some limitations. First, this was a retrospective case–control study, information was obtained from medical records, and the strength of the causal relationship was limited. Second, the once daily assessment of POD in this study might increase the missed diagnosis and underestimate the incidence of POD. Given that, our researchers reviewed the medical records of each patient in detail, found patients whose symptom description matched POD through the chart review, and as supplements only when the two reviewers confirmed the diagnosis in agreement. Moreover, the medical records of perioperative blood loss and fluid replacement were incomplete. We only collected data on intraoperative blood loss, the data on post-operative blood loss, and the perioperative volume of fluid that could not be obtained accurately. Therefore, the relationship between perioperative blood loss and the perioperative fluid volume and POD needs to be verified in further studies. In addition, this was a single-center study. Therefore, the generalizability of the results is limited. Large-scale multicenter studies are needed in the future.

In conclusion, the results of this study showed that pre-operative Hb concentration and perioperative changes in Hb concentration were significantly associated with POD. Age, stroke history, post-operative ICU transfer, and the VAS score in post-operative 2 days were independently associated with POD.