The Potentially Protective Effects of Anti-Lipid, Hypoglycemic, and AntiHypertensive Agents for Perioperative Mortality in Geriatric Group

Background Great efforts were made to collect information and identify risk factors in predicting post-anesthetic mortality. In this study, we use national health insurance data base, including medications, underlying comorbidities and surgical factors to assess the relationship between these factors and mortality after surgery. This is a retrospective, population based study. The study population who underwent general anesthesia (GA) were retrieved from the National Health Insurance Research Database in Taiwan between January 1, 2005 and December 31, 2013. We classied the study patients into 4 major comparison groups by surgical procedures, including major organ transplantation (heart, liver, lung, kidney, or pancreas), CV surgery, major neurosurgery, and others according to the diagnostic codes of the international classication of diseases, ninth revision, clinical modication (ICD-9-CM) codes. We proposed a logistic regression model with valuable variables which can signicantly predicts the post-anesthesia mortality. We also designed different models for 4 subgroups according the results.


Introduction
For subgroup analysis, variables of the underlying comorbidities were almost positively related to post-operation mortality (Table 2). Unexpectedly, hypertension was not associated with increasing risk of post-operation mortality (adjusted OR: 0.77, CI: 0.64-0.93) as other comorbidities were (Table 2).
Nevertheless, prescription with statins usage seemed to be protective for the patient accepted GA (adjusted OR: 0.8, CI: 0.67-0.96, Table 2).
The results of factors associated with post-operation mortality for different surgery types in preliminary group were showed in Table 3 (in-hospital mortality and 30-days mortality). As different surgical types were considered, age was still a signi cantly determining factor associated with post-operation mortality (Table 3). For patients accepting CV surgery, the comorbidity of "valvular heart disease", or "hypertension" was not associated with increasing risk of post-operation mortality ( Table 3). Medication with beta-blockers or statins usage seemed to be protective for the patients undergoing CV operation (Table 3). For the neurosurgery, age and underlying comorbidities were important covariates for the post-operation mortality ( Table 3). As major organ transplantation was considered, the hospital level and the underlying comorbidities might determine the survival rate of the patients after operation. Unsurprisingly, patients accepted organ transplantation at the medical center got better outcomes (Table 3).
For the less risk surgeries, increasing age was associated with a signi cantly higher risk for post-operation mortality. Interesting, hypertension and prescription of thiazolidinediones (TZD) for diabetes mellitus (DM) were with decreasing post-operation mortality (Table 3) in these less risk surgeries. We also observed "valvular heart disease" was associated with increasing post-operation mortality among less risk surgeries, and that phenomenon was converse among the category of CV surgery ( Table 3).
The prediction model was established according to the result of logistic regression model with forward selection of covariates and formulated as coe cient of each risk factor as e- Table 5 and e- Table 6 in the supplement showed. By applying individual's parameters into the formula, it can be used to estimate the probability of 30 -day mortality and in-hospital mortality of each patient.
The ROC analysis of this prediction model was built in preliminary group initially. (Figure 2A and 2B) The optimal AUROC generated from the largest sensitivity and speci city summation for the prediction model of mortality after operation under GA was 0.8746 for predicating in-hospital mortality ( Figure 2A) and 0.8698 for 30-days mortality ( Figure 2B). After the prediction model was built from the preliminary group, further evaluating of the accuracy was conducted in the validation group. The prediction model based on the preliminary group also performed well in the validation group for post-operation mortality.
(AUROC=0.8753 for in-hospital mortality; AUROC= 0.8767 for 30-days mortality) ( Figure 2C and 2D). Validation of this prediction model demonstrated a high level of sensitivity and reliability.

Discussion
To our knowledge, this is the rst and largest nationwide study that formally integrates the individual patient's clinical information (e.g., age, comorbid disease, and medication) to estimate the postoperative mortality. We subdivided surgery into high risk surgery (organ transplantation, CV, and neurosurgery) and nonhigh risk surgery (others) to explore the different risk factors of 30-day and in-hospital mortality in each group. The most important factors that attributed to post-operative mortality are the selected surgery types. Some factors could increase or lower the risk of mortality in non-high risk surgery. Our prediction model built from a smaller set of study population turned out to remain high reliability when applied to the whole population.
Our results showed that age was an important determinant associated with post-operation mortality among different surgical types, also multidimensional score of elderly patients which could possibly improve the prediction was not present in our current database, perioperative prescription could reduce risks of operation. Our data presented the potentially protective effects of anti-lipid, hypoglycemic, and anti-hypertensive agents were encouraging in geriatric preoperative group. It should be part of the patient preoperative preparation as a major contributory factor to the primary causes of perioperative mortality. [9] Hypertension was one of the most common comorbid diseases among the patients who underwent major surgery. However, hypertension was not independently associated with an increased risk of post-operation mortality. As in the previous studies, hypertension was highly prevalent in patients presenting for surgery but its impact on surgical outcome was still under debating. [10] In our study, antihypertensive agents such as beta-blockers and calcium channel blockers were the most commonly prescribed among all of the patients, especially for those in the high-risk groups. On the other hand, valvular heart disease was a major part of cardiovascular surgery. Its presence remained an increased risk for post-operation mortality when the patients underwent less risky surgeries.
Risk factors for the patients undergoing operation with GA are modi able or non-modi able. Regular medication usage might be important to reduce risks of adverse events after operation with GA. In our prediction model, the use of statins, TZD, and beta-blockers for patients was all related to less operative risk and better operation outcomes, indicating that an adequate control of cardiovascular risk factors [11] could decrease overall mortality. [12,13] However, this is the rst study which included medication as a parameter to predict post-anesthetic mortality. The potentially protective effects of anti-lipid, hypoglycemic, and anti-hypertensive agents were encouraging. Further survey with more complete dataset should be warranted.
The types of surgery have determined the risk of post-anesthetic mortality. The major surgeries, such as those involving heart, major vessels, brain, and transplantation, were intrinsically associated with a signi cantly higher risk for mortality. Among these, neurosurgery was associated with the highest risk of post-operation mortality. Besides the surgical types, the prediction model we developed also included a variety of personalized information such as, age, comorbidities, medications, hospital level, and recent medical resource utilization. We might incorporate this tool into an on-site information system, with which when a new patient is to be admitted to the hospital to receive surgery under GA, the individual pro le can be acquired immediately. According to the formula we established (see Appendix), individual's risk of in-hospital and 30-day mortality after general anesthesia can be estimated based on different type of surgery. Clinicians might be able to use this model as a reference to explain the risk of post-anesthetic mortality to patients and their family.

Strengths
This is a nationwide, large-scale study which extract data from National Taiwan Insurance Database which enrolled more than 68,000 surgical cases with GA.
This database is pioneer and unique in the Asian population. The prediction model included a variety of personalized information in addition to surgical types per se. In addition, the use of both a training set to develop a model and a separated validation set has ensured that the model is not only accurate but also stable. The multidimensional and comprehensive prediction model can be applied easily in a real-time clinical setting.

Limitations
There are several limitations in our study. First, the study used a claims database in which some crucial information, such as anthropometric data, blood pressure measuring, laboratory results, is lacking. Using some surrogate variables to develop a prediction model could achieve a differential capacity as high as 88%. But it is unlikely to get even higher with current data. Second, the database lacks some information to determine the status class in the ASA classi cation. We were not able to examine the consistency between our prediction model and ASA classi cation. Third, the patients were exclusively the people in Taiwan. The generalizability to other people remains uncertain.
The results of this study have not only identi ed the risk factors associated with different types of surgeries under general anesthesia, but also explored protective effect on adequate control of chronic disease. We established a clinically applicable prediction model from one population, and further proved its reliability and stability from the other. It is expected that applying this prediction model into clinical practice could improve surgical risk strati cation and further improve patient outcomes.

Consent for publication
There are no con icts of interest to report regarding this study.

Availability of data and materials
All data generated or analyzed during this study are included in this published article as its supplementary les.

Competing interests
There are no con icts of interest to report regarding this study.  supplementarytable.docx