Establishment of risk model for elderly CAP at different age stages: a single-center retrospective observational study

Community-acquired pneumonia (CAP) is one of the main reasons of mortality and morbidity in elderly population, causing substantial clinical and economic impacts. However, clinically available score systems have been shown to demonstrate poor prediction of mortality for patients aged over 65. Especially, no existing clinical model can predict morbidity and mortality for CAP patients among different age stages. Here, we aimed to understand the impact of age variable on the establishment of assessment model and explored prognostic factors and new biomarkers in predicting mortality. We retrospectively analyzed elderly patients with CAP in Minhang Hospital, Fudan University. We used univariate and multiple logistic regression analyses to study the prognostic factors of mortality in each age-based subgroup. The prediction accuracy of the prognostic factors was determined by the Receiver Operating Characteristic curves and the area under the curves. Combination models were established using several logistic regressions to save the predicted probabilities. Four factors with independently prognostic significance were shared among all the groups, namely Albumin, BUN, NLR and Pulse, using univariate analysis and multiple logistic regression analysis. Then we built a model with these 4 variables (as ABNP model) to predict the in-hospital mortality in all three groups. The AUC value of the ABNP model were 0.888 (95% CI 0.854–0.917, p < 0.000), 0.912 (95% CI 0.880–0.938, p < 0.000) and 0.872 (95% CI 0.833–0.905, p < 0.000) in group 1, 2 and 3, respectively. We established a predictive model for mortality based on an age variable -specific study of elderly patients with CAP, with higher AUC value than PSI, CURB-65 and qSOFA in predicting mortality in different age groups (66–75/ 76–85/ over 85 years).

www.nature.com/scientificreports/into three groups classified by age variable: group 1 (aged 66-75 years group with 415 patients, group 2 (aged 76-85 years group with 394 patients, and group 3 (aged over 85 years group containing 365 patients. In these three age-based groups classified above, we divided each group into two populations based on clinical outcome, namely survivor group and non-survivor group.The mortality rate was 11.81% (49/415) in group 118.53% (73/394) in group 2, and 29.59% (108/365) in group 3, respectively (Fig. 1).Next, we analyzed clinically related factors and found out that different variables showed significantly different impacts between survivor group and non-survivor group across the groups classified by age intervals.Specifically, the variables in group 1 included gender, pulse, systolic pressure, leucocyte count, neutrophils count, lymphocyte count, neutrophil-tolymphocyte ratio (NLR), c-reactive protein (CRP), procalcitonin(pct), albumin, urea nitrogen (BUN), D-dimer and cancer(p < 0.05).But in group 2, the statistically significant parameters, which were associated with the prognosis of CAP, contained age, pulse, systolic pressure, diastolic pressure, respiratory rate, leucocyte count, neutrophils count, lymphocyte count, NLR, CRP, pct, albumin, prealbumin, BUN, D-dimer and electrolyte disturbance.Regarding group 3, the variables included pulse, systolic pressure, respiratory rate, leucocyte count, lymphocyte count, NLR, CRP, pct, albumin, low-density lipoprotein, BUN, D-dimer and comorbidities (electrolyte disturbance, cancer, chronic kidney disease, congestive heart failure, coronary heart disease, hypertension) (p < 0.05) (Table 1).Collectively, our results demonstrated that factors showing prognostic values vary among different age groups of the elderly CAP patients.

Prediction of mortality by ROC curves.
Based on above finding, we believe it is reasonable to build a model by these 4 variables (as ABNP model) to predict the in-hospital mortality across all three groups.Further analysis showed that the AUC values of the ABNP model were 0.888 (95% CI 0.854-0.917,p < 0.000), 0.912 (95% CI 0.880-0.938,p < 0.000) and 0.872 (95% CI 0.833-0.905,p < 0.000) in group 1, 2 and 3, respectively (Table 3).As a comparison, we also calculated the AUCs of CURB-65, PSI and qSOFA in all the groups (Table 3) and compared them with the value from ABNP model in predicting in-hospital mortality (Table 4, Figs. 3, 4, 5).Interestingly, the results showed that ABNP model showed superior predictive efficiency when compared to CURB-65 (AUC = 0.827, p = 0.049)/ (AUC = 0.863, p = 0.008), PSI (AUC = 0.821, p = 0.045)/ (AUC = 0.863, p = 0.040) and qSOFA (AUC = 0.766, p = 0.004)/ (AUC = 0.773, p < 0.000) in both group 1 and 2, respectively.Moreover, in group 3, even though no significant difference was observed between ABNP model and PSI (AUC = 0.860, p = 0.060), our new established model (ABNP model) still showed better performance than CURB-65 (AUC = 0.809, p = 0.009) Consistent with previous studies, the mortality rates were 11.81%,18.53%and 29.59% in three groups, respectively, which showed the more pronounced mortality rate with increased age 19,20 .The highest mortality in CAP aged over 85 years group can be explained by the death-associated comorbidities.As shown in Table 1, we observed that comorbidities showed significantly difference between survivors and non-survivors in this group, www.nature.com/scientificreports/including electrolyte disturbance, cancer, chronic kidney disease, congestive heart failure, coronary heart disease and hypertension.Moreover, several factors, like cancer, chronic kidney disease and hypertension, were still independent prognostic factors after the multiple logistic analysis.However, no comorbidities were significantly association with mortality in younger cohorts.This observation is accordant with previous studies of ours and others.For instance, we previously revealed that comorbidities were independent risk factors influencing inhospital mortality in patients over 80 years old with CAP 21 .As well, Ghia et al. found comorbid conditions like chronic obstructive pulmonary disease, hypertension were common risk factors for CAP in the Indian population 22 .On this basis, we propose that more attention should be paid to the care of comorbidities in elderly patients with CAP, especially aged over 85.Another interesting finding is that there are four variables (albumin, BUN, NLR and pulse) independently influencing prognosis in all three age-based subgroups.These variables have been studied and used in clinic.BUN and Pulse have been applied to some assessment scores, such as CURB-65 and PSI, and demonstrated to predict the prognosis of patients with CAP [23][24][25][26] .Additionally, although NLR and Albumin do not belong to common assessment score systems of CAP severity such as CURB-65 and PSI, we also found out NLR and Albumin can improve the predictive ability for mortality of elderly CAP, even in age-based elderly subgroups.
NLR, short for the ratio of absolute neutrophil count to absolute lymphocyte count, has also been identified to predict adverse outcome of patients with CAP [27][28][29][30] .Specifically, Cataudella et al. 27 found NLR predicted 30-day mortality and performed better than PSI and CURB-65 score systems.Thirty-day mortality was 30% in those with a NLR between 11.12 and 13.4%, but 50% in those with a NLR between 13.4 and 28.3.Moreover, Feng et al. 31 also discovered NLR was the independent factor influencing in-hospital mortality in elderly patients with CAP and showed higher AUC value than CURB-65 (0.72 vs. 0.678, p < 0.05).Therefore, a growing number of studies emphasize the importance of NLR to improve the ability of predicting adverse outcome in CAP patients when combined with other factors.A nomogram model composed by NLR was established by Lv et al. 32 to predict mortality in elderly patients with CAP, and the AUC of the model was 0.9, which was proved to be superior to CURB-65 and PSI.Collectively, NLR is a simple, easily measured, yet promising marker for Table 2. Multivariate analysis for mortality in three groups.Abbreviations: OR, odds ratio; CI, confidence internal; NLR, neutrophil-lymphocyte ratio; CRP, c-reactive protein; pct, procalcitonin; BUN, blood urea nitrogen.www.nature.com/scientificreports/predicting outcomes in patients with CAP.Its value, either alone or in conjunction with other biomarkers, need to be further investigated.The fourth variable worth attention is the serum albumin.Typically, albumin is well-known for its important roles in immune regulation and antimicrobial 33,34 .Accumulating evidence show that albumin is related to the prognosis of CAP patients.In one study, Sakakibara et al. 35 established a new score model including albumin to predict severe adverse events (including death) in CAP patients, which exhibits a higher AUC value (0.85) compared with the other predictive models.Furthermore, another study by Shirata et al. 25 developed another albumin-based system (using cutoff as 3.0 g/dL) to predict mortality in older patients with CAP, showing a higher AUC (0.809) than that of CURB-65.In addition, albumin decreased with aging for several potential reasons such as decline in cognition, poor oral health, and dysphagia 36,37 .Thus, it is necessary to increase the level of albumin in elderly patients with CAP, which may improve the prognosis.Several methods can be utilized, such as direct infusion of human albumin.Also, the nasogastric feeding was another preferable option to improve the albumin in elderly patients, especially in patients with decline in cognition after stroke, dysphagia and so on 38 .Finally, cumulative studies also support that the nasogastric feeding tube is efficient to deliver nutrients and/ or fluids to the gastrointestinal tract effectively and play a central role in the management of elderly who were malnourished or hypoalbuminemia [39][40][41][42] .

Variables
Notably, the new established model (ABNP model) shows superiority over clinically used tools (CURB-65, PSI and qSOFA) regarding the prediction of mortality.Notably, the AUC of ABNP model was still higher than PSI score even though there is no significant difference between them in the subgroup aged over 85 years.This could possibly be explained by the contribution of comorbidities.Multiple logistic analysis showed that  comorbidities (cancer, chronic-kidney disease and hypertension) were also independent variables influencing mortality in aged over 85 years patients, besides ABNP-associated factors (albumin, BUN, NLR and pulse).Taken as a whole, we arrived a conclusion that ABNP model was an improved scoring system for prognosis prediction in elder CAP patients.There are some limitations.Firstly, this study is a single-center study, which leads to a limited number of samples and may even cause bias in sample collection.Thus, the results of this study should be verified in multicenter, large-sample studies in the future.Secondly, our study is a retrospective observational study.This may cause several issues, including the possible poor quality of available data due to undesigned study, the possible absence of important data on potential confounding factors and differential losses to follow up on study cohort.Therefore, prospective studies, if applicable, are essential to increase the reliability.Thirdly, other clinical factors are not taken into consideration, such as antibiotic therapy and pathogen infection, and some data are missing for individual patients, such as D-dimer and prealbumin.Fourthly, some data analyzed in this study might not show authentication.For instance, odds ratio values of hypertension and cancer in our analysis were inflated.This might be due to a limited number of samples with hypertension (or cancer) in survivor or non-survivor group.This situation may be related to the special conditions in a data set and this is known as "monotone likelihood" 43 .Thus, we will collect more patient data and apply reconstruction of the interval estimation based on profile penalized log likelihood (PPL) to solve this concern 44 .Finally, we did not take functional decline or frailty into account, which could influence the prognosis of patients with CAP in elderly patients [45][46][47] .Thus, more studies with large population need to be designed in the future.

Conclusions
We established an early prediction model based on an age-group-specific study of elderly patients with CAP.The new model of the AUCs in predicting mortality in different age groups (66-75/ 76-85/ over 85 years) were higher than PSI, CURB-65 and qSOFA.

Figure 2 .
Figure 2. Forest plot of multivariate analysis in three age-based groups.The forest plot showed Pulse, NLR, BUN and Albumin were independent factors in group 1; five variables including Pulse, NLR, CRP, Albumin and BUN were demonstrated to be independently and statistically significant in group 2; And Pulse, NLR, Albumin, BUN, Cancer, Chronic-kidney and Hypertension were observed to independently influence the mortality in group 3.

Figure 3 .
Figure 3.The Receiver Operating characteristic (ROC) curves of the four assessment scores for the mortality in group 1.The AUC of ABNP,CURB-65,PSI and qSOFA were 0.888, 0.827, 0.821 and 0.766, respectively.

Figure 4 .
Figure 4.The Receiver Operating characteristic (ROC) curves of the four assessment scores for the mortality in group 2. The AUC of ABNP, CURB-65, PSI and qSOFA were 0.912, 0.863, 0.863 and 0.773, respectively.

Figure 5 .
Figure 5.The Receiver Operating characteristic (ROC) curves of the four assessment scores for the mortality in group 3.The AUC of ABNP, CURB-65, PSI and qSOFA were 0.872, 0.809, 0.860 and 0.728, respectively.

Figure 1. Mortality
rate in three aged-based groups.The mortality rates were 11.81%, 18.53% and 29.59% in three age-based groups, respectively.
subpopulations with CAP.This study, to our best knowledge, is the first study to understand adjusted parameters for prognosis across different aged groups in elderly CAP patients.Here, we divided elderly CAP patients into three groups classified by age and established a new model (ABNP model) based on parameters derived from prognostic values shared by all the age-dependent subgroups.

Table 1 .
Basic characteristics of three age-based group patients.