Prediction of prolonged mechanical ventilation in critically ill obstetric patients: Ten years of data from a tertiary teaching hospital in mainland China

Maternal admission to the intensive care unit (ICU) during pregnancy or in the postpartum period is a marker of severe acute maternal morbidity. Mechanical ventilation is an important and basic method of maintaining life support in the ICU, but prolonged mechanical ventilation (PMV) is associated with a prolonged length of hospital stay and other adverse outcomes. Therefore, we conducted this retrospective study to describe morbidity and further try to identify the risk factors for PMV in critically ill obstetric women. The clinical data were obtained from a single-centre retrospective comparative study of 143 critically ill obstetric patients at a tertiary teaching hospital in mainland China between January 1, 2009, and December 31, 2019. Clinical and obstetric parameters were collected to analyse the risk factors for PMV. Patients were separated into groups with and without PMV. Potential risk factors were identied by univariate testing. Multivariate logistic regression was used to evaluate independent predictors of PMV.


Background
Although maternal deaths in China have decreased substantially in recent decades, the incidence of severe acute maternal morbidity (SAMM) is still high [1,2]. Due to the implementation of the two-child policy in China, since 2016, the number of pregnant women with an older maternal age and history of caesarean section has increased, which has further increased the incidence of gestational diabetes mellitus, dangerous placenta previa, placental implantation, and severe postpartum haemorrhage [3][4][5].
Maternal admission to the intensive care unit (ICU) during pregnancy or in the postpartum period is a marker of SAMM [2,[6][7][8][9][10][11]. ICU admission of pregnant and postpartum women presents signi cant challenges to ICU clinicians because of altered maternal physiology, fetal considerations and medical emergencies associated with pregnancy [12,13]. Mechanical ventilation is an important and basic manner of maintaining life support in the ICU, but prolonged mechanical ventilation (PMV) increases the risk of a prolonged length of hospital stay, increased hospitalization costs, and other adverse outcomes [14,15]. Therefore, exploring the risk factors for prolonged mechanical ventilation in critically ill pregnant and postpartum women is important for quality and prognosis improvement. However, little is known about the proportion and risk factors for PMV in critically ill obstetric patients. Therefore, we conducted this retrospective study to describe morbidity and further try to identify the risk factors for PMV in critically ill obstetric women.

Participants
The clinical data were obtained from a single-centre retrospective comparative study of 143 critically ill obstetric patients at a tertiary teaching hospital in mainland China between January 1, 2009, and December 31, 2019. This study was approved by the institutional review board following the Declaration of Helsinki principles.
The inclusion criteria for this study were obstetric patients (i.e., those who were pregnant or up to 42 days postpartum) with the following conditions: (1) entered the intensive care unit (ICU) and (2) treated with mechanical ventilation (MV). The exclusion criteria were as follows: (1) age < 18 years; (2) those who signed do-not-resuscitate; and (3) death within 24 hours after MV.

Clinical and obstetric parameters
The baseline clinical and obstetric patient characteristics were collected within the rst 24 hours after mechanical ventilation (the worst results within the rst 24 hours were chosen) (see Table 1). Clinical parameters included age, causes for ICU admission, Acute Physiology and Chronic Health Evaluation (APACHE) II score, body mass index (BMI), creatinine, acute kidney injury (AKI), troponin T, myocardial injury, total bilirubin (TBIL), albumin, brain natriuretic peptide (BNP), white blood cell (WBC) count, platelets arterial blood pH, arterial partial pressure of carbon dioxide (PaCO 2 ), the ratio of the arterial partial pressure of oxygen and the fraction of inspired oxygen (PaO 2 /FiO 2 ), and lactate. Acute kidney injury (AKI) was identi ed according to the Kidney Disease: Improving Global Outcomes (KDIGO) de nition as one of the following: 1) An increase in serum creatinine by ≥0.3 mg/dl (≥26.5 µmol/l) within 48 hours; 2) An increase in serum creatinine to ≥1.5 times baseline within the previous 7 days; 3) Urine volume ≤0.5 ml/kg/h for 6 hours [16]. Myocardial injury was de ned as an increase in blood cardiac troponin I with a cut-off value of 34 pg/mL in our hospital [17].
Obstetric parameters included gestational weeks, obstetric causes of ICU admission, and the estimated blood loss volume during delivery.

Outcomes
The outcomes of the critically ill obstetric patients in our study included the length of ICU stay, length of hospital stay, length of mechanical ventilation, and hospital mortality.

Prolonged mechanical ventilation
Previous studies used prede ned values of mechanical ventilation time, ranging from 24 hours to 72 hours, and even to 21 days, to de ne critically ill patients as having PMV [18]. In this study, PMV was de ned as a mechanical ventilation length of more than 24 hours. This de nition was in accordance with some previous studies [19,20] and also guided by the median length of mechanical ventilation of this study. According to PMV status, patients were divided into two groups: the PMV group and the non-PMV group.

Statistical analysis
All the analyses were performed with Stata software, version 15.1 (Stata Corp). We evaluated the normality of the continuous variables and found that all data were abnormally distributed. Therefore, continuous data were expressed as medians with interquartile ranges. Categorical variables are shown as proportions. In the univariate testing, continuous variables were examined using Kruskal-Wallis equalityof-populations rank tests. Categorical variables were examined using Pearson chi-square tests or Fisher exact tests where appropriate. Predictors with a P value of <0.05 on univariate analysis were identi ed as risk factors for PMV. The variance in ation factor (VIF) and tolerance were used to test the multicollinearity of the risk factors. A VIF > 10 or tolerance < 0.1 was identi ed as signi cant multicollinearity. Binary logistic regression was used to determine independent risk factors for PMV. The Hosmer-Lemeshow test was used to estimate the goodness of t for the logistic regression mode. We generated a receiver operating characteristic (ROC) curve using predicted probability values from the logistic regression. The coefplot module was used to plot the regression coe cients. We used a nomogram to demonstrate the risk points and probability of independent risk factors for predicting PMV. A P value of < 0.05 in 2-sided tests was statistically signi cant.

Participants
Out of 29,236 hospital deliveries, 265 critically ill obstetric patients entered the ICU between January 1, 2009, and December 31, 2019, in our hospital. One hundred forty-ve (54.7%) of them were treated with mechanical ventilation. Two were excluded because of death within 24 hours after mechanical ventilation. Finally, one hundred forty-three patients were enrolled in this study (see Fig. 1). Among these patients, 52 (36.4%) were transferred from other hospital and other 91 (63.6%) patients gave birth at our hospital. Seven (4.9%) of those patients were admitted during pregnancy and 136 (95.1%) admitted postpartum.

Prolonged mechanical ventilation and outcomes
Sixty-ve critically ill obstetric patients (45.5%) underwent PMV. The length of ICU stay (P<0.001) and hospital stay (P<0.001) were both signi cantly longer in the PMV group than in the non-PMV group. The mortality was very low, and there was no signi cant difference between the two groups (P=0.229).

Univariate analysis
Compared with non-PMV patients, PMV patients had a higher APACHE II score (P=0.008) and a larger amount of estimated blood loss during delivery (P<0.001). The incidence of AKI (P<0.001) and myocardial injury (P<0.001) was signi cantly higher in the PMV group than in the non-PMV group. PMV patients had signi cantly higher blood creatinine(P<0.001), troponin(P<0.001), TBIL (P<0.001), BNP (P<0.001), PaCO 2 (P=0.012), and lactate (P=0.002) levels. The blood platelet count (P=0.026) and the arterial blood PaO 2 /FiO 2 (P<0.001) were signi cantly lower in the PMV group. No signi cant difference in age, BMI, gestational weeks, causes of ICU admission, blood WBC count, blood albumin level, arterial blood pH, or bicarbonate was observed (see Table 1 and Table 2).

Multivariate analysis
Variables including APACHE II score, BMI, AKI, myocardial injury, TBIL, BNP, platelets, PaCO 2 , PaO 2 /FiO 2 , and estimated blood loss with a P value of <0.05 in the univariate analysis were identi ed as risk factors for PMV. We tested the multicollinearity of all the risk factors. The results showed that the largest VIF was 1.66 and the smallest tolerance was 0.60 (see Additional le 1). Therefore, we included all 10 chosen risk factors in the logistic regression model, and we found that estimated blood loss (L) (odds ratio (OR) =1.296, P=0.029), AKI (OR=4.305, P=0.013), myocardial injury (OR=4.586, P=0.012), and PaO 2 /FiO 2 (mmHg) (OR=0.989, P<0.001) were independent risk factors for PMV in critically ill obstetric patients (see Table 3). The Hosmer-Lemeshow test showed that the t for the logistic regression model was good (P=0.097, chi2=153.56). The ROC curve based on the predicted probability of the logistic regression is shown in Fig. 2, and the area under the curve was 0.934 (95% CI, 0.895 to 0.974). The plot of the regression coe cients is shown in Fig. 3.

Nomogram
Through the logistic regression model, we built a prognostic nomogram incorporating the above independent prognostic factors for visualization and facilitation of clinical practice, as shown in Fig. 4. In this model, we transferred PaO 2 /FiO 2 into four classes based on the Berlin de nition of acute respiratory distress syndrome (ARDS) to simplify clinical practice [21].

Discussion
To our knowledge, this is the rst study to identify the characteristics of mechanical ventilation and the risk factors for PMV in critically ill pregnant and postpartum patients. In our study, the incidence of mechanical ventilation was 54.7% among obstetric patients admitted to the ICU. Estimated blood loss, AKI, myocardial injury, and PaO 2 /FiO 2 were independent risk factors for PMV in critically ill obstetric patients.
In our study, more than half of critically ill obstetric patients received mechanical ventilation, and 45.5% of them had a ventilation time of more than 24 hours. These results are similar to those of another multicentre report. Zhao et al. enrolled 491 obstetric patients with ICU admissions from three tertiary hospitals in China. They found that 40% of the patients underwent intubation and mechanical ventilation, and the median length of mechanical ventilation was one day [2].
Our study showed that estimated blood loss was an independent risk factor for the development of PMV. Several studies have shown that the most common cause of pregnancy-related admission to the ICU is obstetric haemorrhage [2,12]. A multicentre study in China found that postpartum haemorrhage pregnancy-related ICU admissions and showed that the leading cause of transfer to ICUs was obstetric haemorrhage (4,043; 34.2%) [12]. We also found that postpartum haemorrhage was the main reason for ICU admission (87/143; 35%). Possible reasons for the association of estimated blood loss and PMV included the following: massive obstetric haemorrhage that induced hypovolemic shock and tissue hypoperfusion and further caused lung injury; an independent, dose-dependent relationship between blood transfusion and the subsequent development of acute lung injury; and massive haemorrhage and subsequent resuscitation leading to uid overload, which is associated with pulmonary oedema and ventilator dependence [22,23]. Therefore, timely treatment of postpartum haemorrhage (e.g., surgical procedures including uterine artery ligation, uterine packing, and hysterectomy) could potentially prevent lung injury and reduce the length of mechanical ventilation.
In our study, obstetric patients with an AKI were more likely to have PMV than those without an AKI. Ghauri et al. systematically reviewed the predictors of the need for PMV in adult patients admitted to ICUs for medical and surgical needs. They found that kidney dysfunction was one of the most signi cant independent predictors of the need for PMV [18]. Clark et al. retrospectively assessed 130 ICU patients and showed that serum creatinine levels greater than 2.0 mg/dL were independently associated with PMV. They further validated their ndings in a prospective trial. Acute kidney injury is one of the manifestations of insu cient organ perfusion [24]. On the other hand, kidney dysfunction can cause uid accumulation and tissue oedema. Therefore, it is very important to monitor renal function and treat potential causes in critically ill obstetric patients in a timely manner.
Our results demonstrated that myocardial injury was an independent risk factor for the development of PMV. We de ned myocardial injury as an increase in blood cardiac troponin I [17]. Myocardial injury may be caused by hypoperfusion, hypoxia, global ischaemia, surgery, and sepsis in noncardiac patients. Karpati et al. enrolled 55 parturients with severe postpartum haemorrhage and haemorrhagic shock and found that 51% of them had myocardial injury (elevated serum levels of cardiac troponin I). They also identi ed low systolic and diastolic arterial blood pressure (< 88 and < 50 mmHg, respectively) and increased heart rate (> 115 beats/min) as independent predictors of myocardial injury [25]. A systematic review included nine studies involving 719 patients who suffer from pre-eclampsia and concluded that in such pre-eclampsia pregnant women cardiac troponin I might be elevated [26]. Previous studies also showed that myocardial injury was associated with mortality among patients undergoing noncardiac surgery [17,[27][28]. Our previous study retrospectively enrolled 381 critically ill patients who underwent major abdominal surgery. We found that myocardial injury is an independent risk factor for weaning failure from mechanical ventilation [29]. Abdalla et al. also demonstrated that troponin elevation was associated with a longer duration of mechanical ventilation in patients who were admitted to the ICU with sepsis [27].
According to our logistic regression analysis, PaO 2 /FiO 2 is an independent risk factor for the development of PMV. PaO 2 /FiO 2 is an important variable for evaluating lung injury, and it is the basis for the Berlin de nition of ARDS risk strati cation [21]. PaO 2 /FiO 2 is also an important indicator for evaluating mechanical ventilation weaning [30]. Therefore, we need to monitor PaO 2 /FiO 2 to evaluate the lung function of critically ill obstetric patients and improve PaO 2 /FiO 2 through mechanical ventilation and other aetiological treatments.
We found that not only the pulmonary function parameter PaO 2 /FiO 2 was a risk factor for PMV, but also estimated blood loss, AKI, and myocardial injury were independent risk factors. Therefore, the following measures are crucially important in clinical practice so as to shorten the length of mechanical ventilation: 1) early recognize and timely control the obstetric haemorrhage; 2) maintain hemodynamic stability and tissue perfusion to prevent organ dysfunction; 3) closely monitor the function of kidney, heart, and other organs in critically ill obstetric patients; 4) use lung-protective ventilation in lung injury patients (PaO 2 /FiO 2 <300mmHg).
This study has several limitations. First, the data were retrospectively collected. We could not control for some variables, which may have resulted in data bias. Second, the study was a single-centre study, and most obstetric patients were admitted to the ICU based on local criteria and local practice. Third, we did not distinguish the patients between pregnant and postpartum phase. Fourth, we did not analysis the data or the potential in uence of transfusion of blood products. Fifth, most of the clinical parameters we included were collected after admission to ICU, so this study may have delay when recognizing the deterioration of patients which might have existed prior to ICU admission. Sixth, we derived four independent risk factors for prolonged mechanical ventilation and built a nomogram for visualization and facilitation of clinical practice. However, we did not validate the nomogram with a new database. A prospective, multicentre study is needed to address these issues and validate our ndings.

Conclusions
This retrospective study shows that the independent risk factors for PMV in critically ill obstetric patients are estimated blood loss, AKI, myocardial injury, and

39.2)
Abbreviations: IQR interquartile ranges, PMV prolonged mechanical ventilation, BMI body mass index, APACHE acute physiology and chronic health evaluation, ICU intensive care unit, AKI acute kidney injury, BNP brain natriuretic peptide, WBC white blood cell, TBIL total bilirubin, no. Number, PaCO 2 arterial partial pressure of carbon dioxide, PaO 2 /FiO 2 the ratio of the arterial partial pressure of oxygen and the fraction of inspired oxygen.  Abbreviations: AKI acute kidney injury, PaO 2 /FiO 2 the ratio of the arterial partial pressure of oxygen and the fraction of inspired oxygen. Figure 1 Algorithm for selection of critically ill obstetric patients with mechanical ventilation.

Figure 2
The ROC curve using predicted probability values from the logistic regression. The area under the curve was 0.934, 95% CI was 0.895 to 0.974.