Association between serum calcium and prognosis in patients with acute pulmonary embolism and the optimization of pulmonary embolism severity index

This is a single-center, retrospective, observational study. Consecutive inpatients enrolled met the inclusion criteria: aged 18 years or older, diagnosed with PTE in Peking Union Medical College Hospital (PUMCH) from January 1, 2012 to January 18, 2019. Patients were excluded if they were diagnosed and transferred from other health care facility due to lack of initial clinical data. Patients without important follow-up data were also excluded for the relevant mortality analysis. The diagnosis of PTE required to be confirmed with computed tomographic pulmonary angiography (CTPA), enhanced computed tomography of chest, scintigraphic ventilation-perfusion (V/Q) scan revealing high probability of PTE or to be diagnosed clinically by qualified specialist based on patients’ typical symptoms of PTE, finding deep venous thrombosis (DVT) in extremity by venous ultrasound/phlebography and positive D-dimer.

This study was approved by the Institutional Review Board of Peking Union Medical College Hospital (Ethical review number: B164), in accordance with the Declaration of Helsinki and also registered the clinical trial with identifiers of NCT04411888.

Eligible patients were searching according to the diagnosis code of PTE (ICD-Code: 126) in the hospital electronic medical record system. Patients identification and risk factors related to the prognosis of PTE were collected by three qualified doctors through the hospital electronic medical record system and reviewed by a specialist from pneumology department. The risk factors comprised variables in PESI, that is, age, gender, body temperature, pulse rate, respiratory rate, blood pressure, cancer, chronic heart failure, chronic pulmonary disease, altered mental status and arterial oxyhaemoglobin saturation, and other clinical characteristics, admission laboratory tests, imaging examinations, including dyspnea, chest pain, hemoptysis, syncope, surgery or trauma within 3 months, immobilization state, previous history of DVT, cardiopulmonary resuscitation (CPR) in hospital, glucocorticoid therapy history, hyperlipidemia or diabetes, hypertension, white blood cell, neutrophil proportion, hemoglobin, platelet, D-dimer, alanine aminotransferase (ALT), glutamyl transferase (GGT), albumin, creatinine, creatine kinase-Mb (CKMB), cardiac troponin I (cTnI), N-terminal B-type natriuretic peptide (NT-proBNP), serum calcium, serum potassium, serum sodium, serum chlorine, blood glucose, pH value, echocardiography et al. All the patients enrolled completed at least 1-month follow-up and confirmed the survival status as of January 18, 2019. The follow-up data was ascertained by interviewing patients, families or their physicians by means of telephone.

Continuous and integer variables were presented as the mean value and SD for normally distributed variables, and the median and quartile for abnormal distributed variables. Categorical variables were expressed as their counts and proportions. Univariate logistic regression analysis was performed for variables in PESI and other admission laboratory indicators to assess the association between each factor and 7-day, 14-day, and 30-day PTE mortality respectively. Besides, part of continuous variables which were statistically significantly associated with the PTE mortality, were converted into categorical variables by selecting the maximum Youden index in receiver operating characteristic (ROC) curve as the cutoff value. Then associations between the newly generated categorical variables and PTE mortalities were analyzed.

Then with 30-day all-cause mortality as the outcome, risk factors significantly associated with 30-day PTE mortality were included to develop a multivariate prognosis prediction rule. Notably considering the practicability, the categorical variables generated by continuous variables were included rather than continuous variables themselves. Each variable was assigned an integral point ranging from 0 to 5, and sum of all variables points was calculated to assess patient risk of death. Multiple prediction rules were generated by modifying variable points according to their significance. The rule with maximum area under curve (AUC) value was selected as the optimal prediction rule. By analyzing the ROC curve, the cutoff level of the optimal prediction rule was determined to identify low risk patients. To guarantee the simplicity and applicability, the building process tried to remove less effective variables with relatively lower points, without affecting the sensitivity and specificity of the prediction rule. More details of the building process could be found in the Additional file 3.

Sample size for multivariate prognosis prediction rule was estimated using tool from the website (https://​mvansmeden.​shinyapps.​io/​BeyondEPV/​) [15] recommended by Riley et al. [16]. Validity of the optimal prediction rule, PESI, and sPESI was compared by computing their AUC, prevalence of low-risk patients, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (PLHR), and negative likelihood ratio (NLHR) [17]. The threshold of P-value was set to 0.05. Statistical analysis was conducted with IBM SPSS software (version 26.0, Inc, Chicago, IL, USA), Python 3.7.4, and RStudio 1.1.447.

This study cohort comprised 496 patients with a median age of 61.5 (51, 70) years, and included 229 (46.17%) male patients. PTE was diagnosed in 440 patients (88.71%) by CTPA, V/Q scan in 35 (7.06%), enhanced computed tomography in 9 (1.81%) and clinical diagnosed in 12 (2.42%) patients. As of January 18, 2019, 361 patients survived and 135 died. The median survival time after diagnosis was 644.5 (218, 1693) days. The follow-up period of four patients were less than 30 days (1, 5, 11, and 20 days).

13 (2.62%) patients were excluded because of an undocumented serum calcium and a total of 483 patients were included for the following statistical analysis. Serum calcium levels ranged from 1.48 mmol/L to 2.60 mmol/L with a median value of 2.14 (2.04, 2.23) mmol/L. Univariate logistic regression analysis showed that serum calcium level was significantly associated with 7-day (β: 0.005, 95% CI 0.000–0063, P < 0.001), 14-day (β: 0.010, 95% CI 0.0001–0.087, P < 0.001), and 30-day mortalities (β: 0.035, 95% CI 0.0005–0.230, P = 0.001). Further by selecting 2.13 mmol/L as the cutoff value which had the maximum Youden index in ROC curve (Additional file 1: Fig. S1), patients with or without hypocalcemia were classified into two groups.

Overall, 239 (49.48%) patients with hypocalcemia (serum calcium level ≤ 2.13 mmol/L) showed significant higher 7-day (6.28% vs 2.06%, P = 0.021), 14-day (9.66% vs 2.89%, P = 0.002), 30-day (13.03% vs 4.98%, P = 0.002) mortalities. Adjusting for variables in PESI, cox regression analysis further revealed that hypocalcemia was an independent predictor of 30-day mortality (P = 0.014).

The clinical characteristics of the two groups were presented in Table 1. Distributions of age, gender, clinical symptoms on admission and most of important comorbidities were similar in two groups. Whereas, patients admitted with hypocalcemia had lower systolic and diastolic blood pressure (P = 0.031 and P = 0.036), faster pulse rate and respiratory rate (P = 0.012 and P < 0.001), higher body temperature (P < 0.001), higher shock index (P = 0.004) and more likely to have CPR in hospital (P = 0.014). As for laboratory parameters, patients with hypocalcemia had higher level of neutrophil proportion (P < 0.001), D-dimer (P < 0.001), CKMB (P = 0.005), cTnI (P = 0.046), NT-proBNP (P < 0.001), serum chlorine (P = 0.020), blood glucose (P < 0.001), GGT (P = 0.012) and lower level of hemoglobin (P < 0.001), platelet (P < 0.001), albumin (P < 0.001) and serum potassium (P = 0.031). Echocardiographic parameters were similar between two groups. Besides, patients with hypocalcemia had significantly higher PESI scores than the other group (P = 0.026), but there was no significant difference on their sPESI scores (P = 0.101).

Other univariate associations between categorical variables with 7-day, 14-day, and 30-day mortality were shown in Table 2. For variables of the PESI or sPESI, 5 variables, including "systolic blood pressure (BP) < 100 mmHg", "pulse rate ≥ 110 beats per minute (b.p.m)", "respiratory rate > 30 breaths per min", "altered mental status", and "chronic heart failure" were all significantly associated with 7-day, 14-day, and 30-day mortality. "Age" was associated with 7-day and 14-day mortality, while "age > 80 years" was associated with 14-day and 30-day mortality. "History of heart failure or chronic pulmonary disease" was associated with 7-day and 30-day mortality. Besides, there was association between "cancer" and 30-day mortality. "SaO₂ < 90%" was only associated with 7-day mortality. Other variables including "male sex", "temperature < 36 °C", and "chronic pulmonary disease" had no significant associations with mortalities. Among other laboratory examination variables, serum calcium, NT-proBNP, D-dimer and blood glucose had significant correlations with 7-day, 14-day, and 30-day mortalities. After converting these four variables into the categorical variables (Additional file 1: Fig. S1), hypocalcemia, high NT-proBNP, high D-dimer were still associated with 7-day, 14-day, and 30-day mortalities, and high blood glucose was associated with 7-day mortality.

12 categorical variables significantly associated with 30-day mortality were used to build the prediction rules (see Additional file 2: Fig. S2 and Additional file 3 for details). Minimally required total sample size was 450 and minimally required events per variable (EPV) was 3 for 12 variables by setting events fraction, or 30-day mortality, to 0.08, which were met in this study. Finally, 6 variables were included into the optimal prediction rule shown in Table 3. In the optimal prediction rule, sum of variables points no less than 4 points was identified as the high risk.

In the whole dataset, ROC curves of PESI, sPESI and the optimal prediction rule were shown in Fig. 1. ROC curve of the optimal prediction rule was close to the PESI and sPESI on the beginning, and then climbed obviously higher than them. Validity of three rules was listed at Table 4. Prevalence of 30-day mortality was 0.089 (95% CI 0.065, 0.118). Generally, mean values of AUC, sensitivity (Sen), specificity (Spec), of the optimal prediction rule (Sen: 0.930, Spec: 0.390, AUC: 0.800) were better than both the PESI (Sen: 0.814, Spec: 0.367, AUC: 0.716) and sPESI (Sen: 0.907, Spec: 0.216, AUC: 0.703), although their 95% CIs might overlap. The optimal rule also had advantages of PPV, NPV, PLHR, and NLHR over PESI and sPESI. Especially, specificity and positive likelihood ratio of the optimal prediction rule were significantly higher than the sPESI.