We conducted a study involving elderly T2DM patients with cerebral infarction. The data for this study was obtained from the MIMIC-IV (Medical Information Mart for Intensive Care IV, version 1.0) database, which was maintained by the Massachusetts Institute of Technology (MIT). This database contained records of more than 70,000 adult patients who were admitted to the Intensive Care Unit (ICU) at Beth Israel Deaconess Medical Center in Boston between 2008 and 2019. As the data utilized in this study was sourced from the publicly available MIMIC-IV database, the requirement for informed consent was waived. One of the authors, SZ, was granted full access to the MIMIC-IV public database and was responsible for extracting the necessary data. The certification number for this data extraction work was 57262685. Our research adhered to the guidelines outlined in the Statement for Strengthening Reports of Observational Studies in Epidemiology (STROBE).

A comprehensive cohort comprising 2,197 elderly T2DM patients with cerebral infarction, who were admitted to the Intensive Care Unit (ICU) for the initial occurrence, was meticulously assembled from the MIMIC-IV public database. Following stringent inclusion and exclusion criteria, the final cohort was refined to encompass 448 patients, ensuring an age criterion surpassing 60 years and an ICU stay duration exceeding 24 h.

Based on a thorough review of the literature and clinical expertise, we incorporated additional variables into the analysis for elderly T2DM patients with cerebral infarction. These variables were deemed as potential confounding factors in determining the outcome of mortality. Their inclusion was warranted due to their collinear relationship with SpO₂. Demographics and Admission Information: age, gender, respiratory rate, temperature, heart rate, systolic blood pressure, diastolic blood pressure, average blood pressure, Saturation of Peripheral Oxygen (SpO₂), and Intensive Care Unit length of stay (LOS-ICU). Laboratory parameters: glucose, hematocrit, hemoglobin, platelet, white blood cell (WBC), anion gap, bicarbonate, Blood Urea Nitrogen (BUN), calcium, chloride, creatinine, potassium, International Normalized Ratio (INR), Prothrombin Time (PT), and Activated Partial Thromboplastin Time (APTT). Comorbidities: Congestive heart failure, Renal disease, Severe liver disease and COPD. Severity scores: Charlson Comorbidity Index, APSIII, SAPSII, OASIS, and GCS. Ventilation types: Invasive ventilation, Non-Invasive Ventilation (NIV), Supplemental oxygen, High flow, and Tracheotomy.

Upon admission to the Intensive Care Unit (ICU), a comprehensive set of indicators was gathered on the 1st day. These indicators encompassed factors such as age, gender, Charlson Comorbidity Index, Acute Physiology Score III (APSIII), Simplified Acute Physiology Score II (SAPSII), Oxford Acute Severity of Illness Score (OASIS), Glasgow Coma Scale (GCS), and the length of stay in the Intensive Care Unit (LOS-ICU). The comorbidities of patients were collected, including Congestive heart failure, Renal disease, Severe liver disease and COPD. Within the initial 24 h following admission to the Intensive Care Unit (ICU), the mean values of several vital indicators were documented as observations. These included respiratory rate, temperature, heart rate, systolic blood pressure, diastolic blood pressure, average blood pressure, and glucose levels. Similarly, within the same 24-h period post-ICU admission, the minimum values of various critical indicators were recorded as observations. These indicators encompassed hematocrit, hemoglobin, platelet count, White Blood Cell (WBC) count, bicarbonate levels, Blood Urea Nitrogen (BUN), calcium, chloride, creatinine, sodium, potassium, International Normalized Ratio (INR), Prothrombin Time (PT), and Activated Partial Thromboplastin Time (APTT). Furthermore, the maximum value of the anion gap was documented as an observation within the initial 24 h following admission to the ICU. In the context of the primary independent variable, SpO₂, measurements were acquired during oxygen therapy after admission to hospital. The oxygen therapy encompassed various modalities, including High Flow Nasal Cannula (HFNC), Invasive Ventilation (InvasiveVent), Non-Invasive Ventilation (NonInvasiveVent), Supplemental Oxygen, and Tracheostomy. The median of SpO₂ readings taken throughout the duration of oxygen therapy was employed as a representative measure of the central tendency of oxygen exposure.

The primary outcome under investigation was 1-year mortality, which was defined as the vital status of patients during the 12-month period following their discharge. The death outcome of the study patients was due to massive cerebral infarction.

Using 1-year survival rate post-discharge as an example, we presented the distribution of baseline data for elderly T2DM patients with cerebral infarction after applying inclusion and exclusion criteria. Categorical data were presented using numerical values accompanied by corresponding percentages, while continuous data were depicted either as mean values with their associated standard deviations or as medians along with the interquartile range (IQR). This particular method offered a balanced representation of the data, capturing both the central tendencies and the spread of continuous variables. To gauge distinctions among continuous variables, the analysis of variance was applied. For categorical variables, the chi-square test was utilized to juxtapose the characteristics of the subjects within the outcome group. This meticulous approach offered a holistic perspective on the baseline characteristics, ensuring a thorough evaluation of the factors influencing 1-year survival rates post-discharge. It served to provide a robust foundation for understanding the dynamics at play in this clinical context.

SpO₂ referred to the percentage of hemoglobin combined with oxygen in the blood to all hemoglobin that could be combined, representing the patient 's oxygen saturation. It was an important respiratory physiological parameter that could determine whether the patient had a hypoxic and hyperoxic state. Therefore, we divided the SpO₂ of patients after inclusion and exclusion into four levels: SpO₂ ≤ 94.5%, 94.5% < SpO₂ ≤ 96.5%, 96.5% < SpO₂ ≤ 98.5% and SpO₂ > 98.5% to distinguish and compare the patient 's blood oxygen saturation level.

A restricted cubic spline analysis was conducted with the aim of ascertaining the existence of a noteworthy non-linear correlation between SpO₂ and 1-year mortality. This analytical approach sought to elucidate whether the risk of mortality displaying either low oxygen saturation or elevated oxygen saturation levels exhibited a substantial escalation.

Cox multivariate regression was harnessed to validate and quantify the insights gleaned from the curve fitting process. Three distinct models were constructed sequentially: Model 1 (not adjusted), Model 2 (adjusted for hematocrit and Charlson Comorbidity Index), and Model 3 (adjusted for Model 2 in addition to respiratory rate and SAPSII). These models were instrumental in computing hazard ratios (HR) along with their accompanying 95 percent confidence intervals (CIs), thus shedding light on the correlation between varying SpO₂ levels and 1-year mortality. This comprehensive analysis allowed us to assess the heightened risk of mortality among patients with diverse oxygen saturation levels, encompassing those with low, normal, high, and notably high levels.

We considered that the relationship between SpO₂ levels and 1-year mortality might be influenced by various factors, including Sex, Age, Concomitant diseases such as Congestive heart failure, Temperature, and LOS-ICU. To investigate potential heterogeneity among subgroups, we conducted Cox subgroup analysis, and the interactions between these subgroups were assessed using the likelihood ratio test. Data analyses were carried out using packages R 4.1.2 software and Free Statistics software versions 1.5. Statistical significance was defined as a P-value < 0.05. This comprehensive approach allowed us to explore how these various factors may impact the relationship between SpO₂ levels and 1-year mortality in this specific patient population.

In the MIMIC-IV public database, we identified and selected 2,197 elderly T2DM patients with cerebral infarction who were admitted to the ICU for the first time. Among this group, 1,606 patients were excluded due to substantial missing information, while 92 patients under the age of 60 and 51 patients with < 24 h of ICU hospitalization were also excluded. To address the missing data, mean interpolation techniques were applied. Ultimately, the final cohort comprised 448 elderly T2DM patients with cerebral infarction (as illustrated in Figure 1, the study patient flowchart).

Table 1 presented a comprehensive summary of the demographic and clinical profiles of the patient cohort following their inclusion and exclusion criteria. The study encompassed 448 participants, with 226 individuals being female, constituting 50.4% of the total sample. The average age across the entire cohort was 74.0 [IQR 68.0-82.0] years, and the mean SpO₂ recorded within 24 h after admission stood at 97.2 [IQR 95.9–98.6]%. Notably, within the follow-up period of 1 year after discharge, 161 patients (35.94%) succumbed to their conditions. It was evident that patients in the non-survival group exhibited distinct characteristics compared to their counterparts in the survival group. These patients tended to be older and presented with more severe illness upon admission [Age 78.0 [IQR 70.0–85.0] years vs. 72.0 [IQR 67.0–80.0] years, p < 0.001; Charlson Comorbidity Index 9.0 [IQR 8.0–11.0] vs. 8.0 [IQR 7.0–10.0], p < 0.001]. Their respiratory rate and heart rate were notably higher, and they had elevated average blood pressure levels when compared to the survival group. Additionally, laboratory investigations revealed that patients in the non-survival group displayed higher values for parameters such as anion gap (AG), blood urea nitrogen (BUN), and prothrombin time (PT) in comparison to the survival group. Furthermore, patients in the non-survival group had a prolonged stay in the Intensive Care Unit, indicating the gravity of their medical condition [LOS-ICU 4.2 [IQR 2.6–9.2] vs. 3.1 [IQR 1.7–5.5], p < 0.001]. Remarkably, a heightened prevalence of congestive heart failure was discerned among patients in the non-survival cohort, suggesting a potential comorbidity that could intricately influence their prognostic outlook. Noteworthy disparities in survival rates were also evident across patients with varying ventilation statuses, predominantly featuring Invasive ventilation and Supplemental oxygen modalities. Intriguingly, those subjected to Invasive ventilation exhibited elevated mortality rates. A paradoxical observation emerged as the non-survival group displayed unexpectedly elevated oxygen saturation levels compared to their surviving counterparts [SpO₂ 97.6 [IQR 96.5–98.9]% vs. 96.8 [IQR 95.7–98.2]%, p < 0.001]. This counterintuitive result underscores the imperative for comprehensive investigation to elucidate the underlying mechanisms and unravel the nuanced clinical implications.

Figure 2 illustrated the relationship between 1-year mortality and the average SpO₂ levels recorded on the 1st day following admission. The findings revealed that low SpO₂ levels ( ≤ 94.5%) are linked to a heightened 1-year risk of mortality. In contrast to conventional wisdom, high SpO₂ levels (>96.5%) also correspond to an elevated 1-year mortality risk. While adjusting for covariates, specifically hematocrit and the Charlson Comorbidity Index, we employed restricted cubic spline analysis to flexibly model the association between SpO₂ levels and 1-year mortality in elderly T2DM patients with cerebral infarction. The results unveiled a U-shaped pattern in the relationship between participants' SpO₂ levels and 1-year mortality, with a statistically significant P-value for non-linearity (0.037). According to the restricted cubic spline analysis, the lowest range of 1-year mortality risk fell between SpO₂ levels of 94.5–96.5%. When SpO₂ levels were below 94.5% or above 96.5%, the fitted curve exhibited a moderately steep U-shaped trend. This observation indicated that both low SpO₂ levels and high SpO₂ levels were associated with an increased 1-year mortality risk among elderly T2DM patients with cerebral infarction. These findings challenged conventional assumptions by highlighting the complex and non-linear nature of the relationship between oxygen saturation and 1-year mortality in this patient population.

Using the Cox regression model, we presented in Table 2 both non-adjusted and adjusted estimates regarding the relationship between SpO₂ levels and 1-year mortality in elderly T2DM patients with cerebral infarction. When SpO₂ was used as a categorical variable, with the reference group being individuals with normal SpO₂ levels (94.5% < SpO₂ ≤ 96.5%), Model 1 (No adjusted) revealed that both low SpO₂ levels ( ≤ 94.5%) and high SpO₂ levels (96.5% < SpO₂ ≤ 98.5%) were associated with an elevated risk of 1-year mortality [HR_l: 1.18, HR_h: 1.77]. Conspicuously, notably high SpO₂ levels (>98.5%) were significantly linked to a higher 1-year mortality risk [HR: 2.25]. Model 2, which adjusted for Hematocrit and the Charlson Comorbidity Index, maintained the associations of low ( ≤ 94.5%) and high SpO₂ levels (96.5% < SpO₂ ≤ 98.5%) with an increased 1-year mortality risk [HR_l: 1.4, HR_h: 1.64]. Notably high SpO₂ levels (>98.5%) also persisted as a significant risk factor [HR: 2.06]. In the subsequent analyses, Model 3, which incorporated additional adjustments for respiratory rate and SAPSII based on the foundation of Model 2, and Model 4, featuring further adjustments for PaO₂/FiO₂ and COPD based on the refined Model 3, yielded results in alignment with the outcomes of Models 1 and 2. Importantly, these supplementary adjustments did not introduce significant modifications to the identified associations between SpO₂ levels and 1-year mortality within this specific population. This robust consistency across Models 1 to 4 underscores the robustness and reliability of the observed correlations.

The analysis demonstrated that both low and high SpO₂ levels were associated with an elevated 1-year mortality risk among elderly T2DM patients with cerebral infarction. Interestingly, the risk of mortality appeared to follow a gradient with notably high SpO₂ levels posing the highest risk, followed by high SpO₂ levels, low SpO₂ levels, and finally, normal SpO₂ levels. This finding provided a unique perspective, emphasizing that notably high oxygen saturation levels might have an adverse impact on the 1-year mortality risk, further emphasizing the non-linear and complex nature of this relationship within the context of this patient population.

The MATLAB software was employed to meticulously calculate daily cumulative mortality rates for patients stratified into four distinct SpO₂ subgroups over a 365-day period post-admission. Leveraging survival time data extracted from the MIMIC-IV database, the outcomes were elegantly visualized in Figure 3 using the plot() function, with detailed magnification within a graph window. Upon meticulous scrutiny and analysis of the graphical representation, a distinct pattern emerged on the 365th day, exhibiting a mortality rate hierarchy: normal SpO₂ levels (94.5% < SpO₂ ≤ 96.5%), low SpO₂ levels ( ≤ 94.5%), high SpO₂ levels (96.5% < SpO₂ ≤ 98.5%), and notably high levels (>98.5%). Remarkably, this observed trend harmonized seamlessly with the findings derived from the Cox multivariate regression analysis.

When conducting subgroup analysis, it was observed that there was no statistically significant interaction between SpO₂ and the grouping variables, including Sex, Age, Congestive heart failure, Temperature and ICU length of stay (LOS-ICU; P-values for interaction were >0.05; as depicted in Figure 4).