大數據決策分析於股骨頸骨折麻醉方式應用

Abstract

研究背景 股骨頸骨折為年老族群常見之骨折型態，由於年長者之脆弱狀態，低能量的傷害即可對該族群中造成股骨頸骨折。雖然對於股骨頸骨折的治療共識為盡早進行外科介入，但由於此族群之多重共病症以及脆弱狀態之影響，外科介入所需的麻醉方式以及術後照護皆有許多複雜因素的考量。股骨頸骨折手術之麻醉方式通常為半身麻醉或是全身麻醉，惟半身麻醉為避免造成脊髓腔血腫，與下肢癱瘓，因此禁忌症包含病患凝血功能異常與服用口服抗凝血劑等。此外年長族群經常受到多重共病症以及其脆弱狀態的影響，以及術後可能需要到接受加護病房照護而延長住院康復日數之可能性，亦影響患者最佳麻醉方式之選擇。目前麻醉醫療對於此脆弱族群運用半身麻醉或全身麻醉何者為佳仍未有定論。 研究目的 本研究主要目的有三 (1)探討影響股骨頸骨折年長族群術後須接受一般病房、加護病房照護，以及死亡之病患層面因素與麻醉風險因素之影響因子； (2)建立以實證為基礎包含前述術後狀態轉移之多階段風險預測模式； (3)依據(1)與(2)發展以病患術風險為基礎之精準麻醉方式評估架構。 材料與方法 本研究納入臺北榮民總醫院自2016年至2020年間股骨頸骨折手術病患，以貝氏網路建立包含術後四種狀態(病房、加護病房、出院、死亡)以及包含病患層級因子、麻醉方式層級因子，以及術後併發症因子之轉移狀態多重影響因素結構。本研究運用寇斯等比風險迴歸模型(Cox proportional hazards regression model，Cox PHREG)評估病患術後於前述四種狀態間的轉移影響因子與轉移風險，並據以建立術後危險分數。運用此術後危險分數結合貝氏離散時間與離散狀態馬可夫迴歸模型(Markov regression model with discrete-state and discrete-time)評估不同風險層級之股骨頸骨折病患於術後狀態間轉移機率建立其轉移機率矩陣。利用此實證風險為基礎之轉移機率矩陣，本研究架構含括多因素之多階段術後風險評估模型作為麻醉方式之實證決策依據。 結果 本研究共納入1605位股骨頸骨折病患建立貝氏網路麻醉決策。由Cox PHREG 模式分析結果顯示，影響術後出院轉移機率的因子包含年齡、麻醉方式、ASA評分、術中出血、肺炎、心臟衰竭、糖尿病、術前血紅素低於10g/d。術後發生肺炎則顯著提高由住院轉加護病房之風險(風險對比值(HR): 12.7，95% CI: 3.73-42.88)，並延後由加護病房轉回病房之可能(HR: 0.31，95% CI: 0.13-0.73)。癌症為術後死亡之顯著風險因子(HR: 3.84，95% CI: 1.19-12.43)。本研究運用此寇斯迴歸分析建立之術後轉移風險分數評估此脆弱族群之臨床結果變化。其中康復出院速率以接受半身麻醉方式且術前為低風險病患最快，手術後一週已有85%的病患出院。而同時間(一周)接受半身麻醉但屬於中風險以及高風險病患出院比例分別為75%和71%。若同為接受半身麻醉但具有肺炎且屬於中風險之病患之病患，則需三週後才能達到74%出院。而具有肺炎且術前評估為高風險之病患族群三周出院比率則僅為57%。 肺炎對於接受全身麻醉患者之出院速率亦有顯著影響，接受全身麻醉中具有肺炎者其出院速率遠低於無肺炎者。如同接受半身麻醉病患，全身麻醉病患其術前風險越高之病患，出院速率越慢。接受全身麻醉病患中，無肺炎且術前風險低、中、高者其術後第七天出院機率分別為81.5%、70.1%，以及55.6%，對於接受全身麻醉且發生肺炎族群其21天出院機率分別為68.7%(術前風險為低)、41.2%(術前風險為中)，以及36.7%(術前風險為高)。 不論麻醉方式為半身或全身麻醉，術後發生肺炎對於病患臨床狀態惡化至需要加護病房治療皆為重要影響因素。而無肺炎的病患則是受到其術前風險影響，此術前風險對於病患術後臨床狀態轉移之影響在接受全身麻醉之病患相較於接受半身麻醉這又更為重要。 結論 本研究運用Cox PHREG 評估病患因素、麻醉術式因素，以及術後是否發生肺炎等多層面因素評估其對於病患術後不同臨床狀態進展之影響並且結合貝氏網絡架構以離散時間離散狀態馬可夫模型建立包含術前風險層級，術後是否發生肺炎之病患多階段臨床狀態進展路徑預測，據以提供此脆弱股骨頸骨折病患最佳麻醉方式選擇之實證依據。

Background Femoral neck fracture (FNF) is a common injury among the elderly population caused by even low-energy falls. Current consensus on FNF treatment is surgery at earliest as soon as possible after diagnosis. To reach this goal, two types of anesthesia for FNF surgery, general (GA) and spinal anesthesia (SA) can be selected. Contraindications, such as coagulopathy, under anticoagulant or antiplatelet treatment, may lead to spinal hematoma and paralysis following SA. The elder population are also suffered from comorbidities and associated physiological limitations, which may increase the perioperative risk of unplanned ICU (intensive care unit) admission and death and thus render the type of anesthesia crucial for an optimal recovery after surgical intervention for FNF. Aims The aims of my thesis are (1)to explore the multifaced factors in terms of patient-level and types of anesthesia in associate with the perioperative progression of multiple clinical outcomes including ward, ICU care, recovery and discharge, and death for FNF patients; (2)to construct a risk-based multistate transition across the clinical outcomes stemming from the state-specific risks elucidated in (1); and (3)to establish a framework for precision decision for types of anesthesia on the basis of (1) and (2). Materials and Methods Data on patients receiving FNF surgery between January 1, 2016 and February 29, 2020 were collected from Taipei Veterans General Hospital. Under the framework of the Bayesian Network approach, multifaceted factors including patient-level, types and risks of anesthesiology, and post-operative complications on the risk of the evolution across multiple clinical outcomes (ward care, ICU care, death, and recovery and discharge) were constructed. A Cox proportional hazards regression model (Cox PHREG) was first applied to evaluated the impact of the multifaced factors on the clinical outcomes mentioned above. On the basis of the risk scores estimated from Cox PHREG, a Markov regression model with discrete-state and discrete-time was used to assess their effect on the probabilities of state-specific transitions. Supported by the data driven estimates on the probability transition matrix across defied clinical outcomes, a framework for optimal decision on the types of anesthesiology was then established. Result A total of 1605 subjects receiving FNF surgery were enrolled in this study. From the results of Cox PHREG analysis, factors associated with the risk of delayed discharge include older age, general anesthesia, higher ASA score, blood loss of more than 500ml, heart failure, pneumonia, diabetes, and a preoperative hemoglobin less than 10mg/dl. Among these factors, pneumonia increased the risk of being transferred to ICU (hazard ratio, HR: 12.7，95% CI: 3.73-42.88) and prolonged ICU stay by decreasing the force of transition from ICU to ward (HR: 0.31，95% CI: 0.13-0.73). The comorbidity of cancer increase the risk of in-hospital mortality by 4 fold (HR: 3.84，95% CI: 1.19-12.43). Supported by these quantitative results derived from Cox PHREG, we further construct the transition across the four clinical outcomes defined above by using a discrete-time and discrete-state Markov regression model. Patients with low preoperative risk under SA have the lowest length of stay (LOS). The probability of discharge on the 7th postoperative day for patients under SA and GA without pneumonia were 81% and 85% (low preoperative risk), 75% and 70% (moderate preoperative risk), and 71% and 55% (high preoperative risk), respectively. For subjects receiving SA but with pneumonia and high preoperative risk, it takes three weeks for these subpopulation to reach 74% probability of recovery and discharge, which is 57% among the subpopulation with high preoperative risk. There is significant impact of pneumonia on the chance of discharge among those receiving GA. Consistent with those receive SA, pre-operative risk exerts a significant impact on the chance of discharge for this subpopulation. For the subpopulation receiving GA without pneumonia, the chance of discharge was 81.5%, 70.1%, 55.6% at 7th day for those at low, medium, and high pre-operative risk, respectively. For this subpopulation with pneumonia, the corresponding figures at 21th day were 68.7% (low risk), 41.2% (medium risk), and 36.7% (high risk), respectively. Higher preoperative risk is related to higher chance of ICU admission and prolonged LOS in ICU. Patients with pneumonia had much longer LOS and ICU stay, especially in GA. Discussion Stemming Cox PHREG, a series of risk factors including patient-level, types of anesthesiology, and pneumonia after surgery were incorporated to establish a multifaceted risk score. By using this risk score in conjunction with a Markov regression model in discrete-state and discrete-time with the Bayesian network, we construct the trajectory of the transition across the clinical outcomes of ward care, ICU care, recovery and discharge, and death, which further support an optimal decision on the type of anesthesiology.