Study design and setting:
National Health Insurance registration database (NHIRD)
The Taiwan National Health Insurance (NHI) program is mandatory and universal, offering comprehensive medical care coverage to more than 99% of the country’s population of 23 million people. This nationwide compulsory healthcare program that covers outpatient visits, hospital admissions, prescriptions, interventional procedures, disease profiles, and vital statuses. Taiwan’s NHI records are regularly inspected, and physicians are subject to statutory regulation.[13-18] The National Health Insurance Research Database (NHIRD) thus represents a comprehensive, high-quality record of healthcare episodes provided to the Taiwanese population.
Selection of Participants:
We enrolled patients aged ≥ 18 years who were admitted because of major trauma and developed AKI-RRT during their index admission as recognized by RRT-related procedure codes and survived to hospital discharge. All diagnoses, including trauma-related injuries and baseline comorbidities, were obtained by the codes of International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM). Those that were related to vehicle-related accidents, defined as any of ICD-9-CM E-Code E800 to E844 included in the diagnosis codes, were aggregated to individuals and were enrolled as study subjects.[15] When more than one accident was encountered by a single person, only the first event was counted for analysis. We identified baseline comorbidities from at least three outpatient visits or one inpatient claim within a one-year antecedent to the index admission with first dialysis. This rule was constructed based on a relatively strict criterion and was well validated with good predictive power.[13-16, 19]
Dialysis patients who underwent renal transplantation, vascular access creation, or peritoneal-dialysis catheter implantation for chronic dialysis and those who had received chronic dialysis prior to the index admissions were excluded. To compare the long-term outcomes of interest, we constructed a comparator group of non-traumatic AKI-RRT patients matched to traumatic AKI-RRT patients in a 3:1 ratio based on age, gender, and propensity scores. (Supplemental Table 1)
As all personal information is de-identified in the research database to protect privacy, no informed consent was required, and this study was deemed exempt from a full ethical review by the institutional review board of the National Taiwan University Hospital (201212021RINC).
Outcomes:
Our primary outcome was long-term all-cause mortality after hospital discharge. The secondary outcomes were 30-day all-cause mortality and de novo end-stage kidney disease (ESKD) defined as the requirement of dialysis for at least three months after hospital discharge. We used a selection period of 90 days to define ESKD because all patients receiving dialysis for more than 90 days in Taiwan can apply to the NHI for catastrophic illness registration cards.[20] Each patient was monitored from the date of discharge and was censored at either death, dialysis, or the end of the study (December 31, 2010), whichever occurred first.
Research variables
We recorded the disease severity and patient condition during index hospitalization. Disease severity and patient condition are estimated by the ICU procedure and complications with acute pulmonary (prolonged mechanical ventilation, re-intubation, acute respiratory distress syndrome, pleural effusion, and chest tube insertion), cardiovascular [hemopericardium, hypovolemic shock, cardiac arrest, heart block, atrial fibrillation, extracorporeal membrane oxygenation (ECMO) and intra-aortic balloon pumping (IABP)] and infectious (pneumonia, urinary tract infection and severe sepsis) and other (delirium, stroke and gastrointestinal bleeding) disorders.[21]
The Charlson comorbidity index [22] was computed using baseline comorbidities. Additional adjustments in these models included control for direct effects from age, gender, and comorbidities that are listed in Table 1.
Analysis
Baseline characteristics were described as the percentages for categorical variables and median with interquartile range (IQR) for continuous variables. Differences between the trauma and non-trauma related groups were compared by the independent t-test or χ2 test where appropriate.
To estimate each patient’s propensity score for traumatic AKI-RRT, we fitted a separate multivariable logistic regression model with the factors predicting vehicle trauma as admitting diagnosis in patients with AKI-RRT during index hospitalization.[23] (further seen in supplementary Table 1). The caliper distance is 0.25 and subjects are matched without replacement in this propensity score matching. The estimated propensity score was also added to adjusting the Cox regression model as a single covariate for controlling selection bias. Multivariable logistic regression models before and after propensity-scored matching were applied to estimate odds ratio (OR) of study outcomes after adjusting all the confounders predicting trauma (supplementary Table 1).
The significance levels for entry and stay were set to 0.15 to be conservative. Then, with the aid of substantive knowledge, the best candidate final logistic model was identified manually by dropping the covariates with p value > 0.05 one at a time until all regression coefficients were significantly different from 0.
All the analyses were conducted with R software, version 2.8.1 (Free Software Foundation, Inc., Boston, MA, USA); competing-risk analysis was performed using Stata/MP version 12 (Stata Corporation). A two-sided p-value < 0.05 was considered to be statistically significant.