Mortality and Cause of Death in Patients With Vertebral Fractures

A Longitudinal Follow-Up Study Using a National Sample Cohort

Hyo Geun Choi, MD; Joon Kyu Lee, MD, PhD; Songyong Sim, PhD; Miyoung Kim, MD, PhD


Spine. 2020;45(5):E280-E287. 

In This Article

Materials and Methods

Study Population and Data Collection

The ethics committee of Hallym University approved the use of these data (approval number 2014-I148). The requirement for written informed consent was waived by the Institutional Review Board.

This national cohort study relies on data from the Korean Health Insurance Review and Assessment Service (HIRA)-National Sample Cohort. The Korean National Health Insurance Service, enrollment in which is mandatory, selects samples directly from the entire population database to prevent non-sampling errors. Approximately 2% of the samples (1 million) were selected from the entire Korean population (50 million); these data were classified into 1476 levels (age [18 categories], sex [2 categories], and income level [41 categories]) using randomized stratified systematic sampling methods via proportional allocation to represent the entire population.[16] The cohort database included (i) personal information, (ii) health insurance claim codes (procedures and prescriptions), (iii) diagnostic codes using the International Classification of Disease, 10th edition (ICD-10), (iv) death records from the Korean National Statistical Office (using the Korean Standard Classification of Disease), (v) socio-economic data (residence and income), and (vi) medical examination data for each participant from 2002 to 2013.

Because all Korean citizens are recognized by a 13-digit resident registration number from birth to death, exact population statistics can be determined using this database. All Korean hospitals and clinics use the 13-digit resident registration number to record individual patients in the medical insurance system. Therefore, the risk of overlapping medical records is minimal, even if a patient relocates to another geographical region. Moreover, all medical treatments in Korea can be tracked. In Korea, a notice of death must legally be delivered to an administrative entity before a funeral can be held. Causes and dates of death are recorded by medical doctors on death certificates.

Participant Selection

From 1,125,691 patients with 114,369,638 medical claim codes, we included 23,026 participants diagnosed with vertebral fracture between 2002 and 2013. Vertebral fracture was defined as fracture of a thoracic vertebra (ICD-10 codes: S220) or a lumbar vertebra (S320). The control participants were extracted from among the 1,102,665 remaining participants with no diagnoses of vertebral fractures between 2002 and 2013.

The vertebral fracture participants were matched 1:4 with the control group, accounting for age, group, sex, income, and region of residence. We set the index date as the date of diagnosis of vertebral fracture; participants in the control group were followed from the same index date as their matched counterparts. The follow-up duration was calculated from the index date to the death date or the last date of study (December 31, 2013), and the study individuals were followed until death or censoring. To prevent selection bias, the control group participants were sorted using a random number and then selected in descending order. It was assumed that the control participants were involved at the same time of each matched participant with vertebral fracture; therefore, control group members who died before the time of the matched vertebral fracture participant's incident were excluded. Vertebral fracture participants for whom a sufficient number of matching participants were not identified were excluded (n = 1267). Finally, 21,759 vertebral fracture participants and 87,036 control participants were analyzed (Figure 1).

Figure 1.

A schematic illustration of the participant selection process used in the present study. Of a total of 1,125,691 participants, 23,026 with vertebral fracture were selected. The vertebral fracture participants were matched 1:4 with a control group in which patients were not diagnosed with vertebral fracture. Finally, 21,759 vertebral fracture and 87,036 control participants were included.


Age groups were classified using 5-year intervals (0–4, 5–9, 10–14…, and 85+ years old); 18 age groups were designated. The income groups were initially divided into 41 classes (1 health aid class, 20 self-employment health insurance classes, and 20 employment health insurance classes) but were re-categorized into 11 classes (class 1 [lowest income] to 11 [highest income]). Regions of residence were divided into 16 areas according to the administrative district and were regrouped into urban (Seoul, Busan, Daegu, Incheon, Gwangju, Daejeon, and Ulsan) and rural (Gyeonggi, Gangwon, Chungcheongbuk, Chungcheongnam, Jeollabuk, Jeollanam, Gyeongsangbuk, Gyeongsangnam, and Jeju) areas.

The causes of death were classified according to the Korean standard classification of diseases, which is developed by the World Health Organization and based on the ICD, into 12 classifications: (i) Infections and parasitic diseases (A00–B99); (ii) Neoplasms (C00–D48); (iii) Metabolic diseases (E00–E90); (iv) Mental and behavioral disorders (F00–F99); (v) Neurologic diseases (G00–G99); (vi) Circulatory diseases (I00–I99); (vii) Respiratory diseases (J00–J99); (viii) Digestive diseases (K00–K93); (ix) Diseases of the musculoskeletal system and connective tissue (M00–M99); (x) Genitourinary disease (N00–N99); (xi) Abnormal finding (not elsewhere classified, R00–R99); (xii) Trauma (injury, poisoning and certain other consequences of external causes, S00–T98); and (xiii) Others (diseases of the blood and blood-forming organs and certain immune system disorders, D50–D89; diseases of the skin and subcutaneous tissue, L00–L99). The Charlson Comorbidity Index (CCI) was used for 17 comorbidities as a continuous variable (0 [no comorbidity] through 29 [multiple comorbidities]).[17]

Statistical Analyses

The chi-square or Fisher exact test was used to compare the general characteristics of participants in the two groups, as well as to compare their mortality rates according to the cause of death. The false discovery rate was used to adjust for the incorrect rejections of the null hypothesis.

To determine hazard ratios (HRs) for mortality as a function of vertebral fracture, a stratified Cox proportional hazards model, both crude (simple) and adjusted for the CCI, was used. Age, sex, income, and region of residence were stratified. Kaplan–Meier analysis and log-rank test were used to show the survival rate through the study period. Two-tailed P-values <0.05 were considered statistically significant. The results were analyzed using the SPSS software, version 21.0 (IBM, Armonk, NY).