Mortality of Midlife Women With Surgically Verified Endometriosis

A Cohort Study Including 2.5 Million Person-years of Observation

L. Saavalainen; A. But; A. Tiitinen; P.Härkki; M. Gissler; J. Haukka; O. Heikinheimo

Disclosures

Hum Reprod. 2019;34(8):1576-1586. 

In This Article

Materials and Methods

Study Population

To identify surgically diagnosed cases, all endometriosis-associated diagnoses (International Classification of Diseases version 9 [ICD-91987-1995]: 6171A, 6172A, 6173A, 6173B, 6174A, 6175A, 6176A, 6178X, and 6179X; version 10 [ICD-101996-2012]: N80.1–N80.6, N80.8, N80.80, N80.81, N80.89, and N80.9), as the main or subsidiary diagnosis, in combination with relevant concomitant surgical codes from 1987 to 2012 (n = 49 956), were identified from the Finnish Hospital Discharge Register (FHDR). The managing clinician sets the primary and secondary ICD codes for each procedure according to their clinical relevance. Adenomyosis as a sole diagnosis was not included. The index day was set to the day of discharge from the first hospital episode fulfilling the definition of surgically verified endometriosis.

There were altogether 57 713 endometriosis diagnoses among 49 956 women in the index procedure. Peritoneal endometriosis was the most common diagnosis (n = 26 299; 46%) followed by ovarian endometriosis (n = 24 343; 42%), other or unknown endometriosis (n = 3578; 6%), and deep infiltrating endometriosis (n = 3493; 6%).

The reference cohort (n = 98 824) was randomly drawn by a computer from the Finnish Population Information System. The reference cohort was first constructed by selecting, for each endometriosis patient, two women who were alive, lived in the same municipality and were of similar age on the index date, and had no surgically verified endometriosis according to the FHDR records over the period of 1987–2012 and no hospital admissions due to endometriosis 1983–1987. The final reference cohort included one to two women for each endometriosis patient fulfilling these criteria.

Data Sources

In Finland administrative and health data for the entire population have been collected for decades using well-established and high-standard procedures (Gissler and Haukka, 2004; Sund, 2012; Pukkala et al., 2018). A unique personal identity code, issued to every resident in Finland since 1964–1968, secures a reliable data recording and allows data linkage since 1969.

The FHDR includes personal identity codes, codes of diseases according to the ICD, and dates for each hospital visit since 1967 from both general and private sector inpatients, as well as for day surgeries from 1994 onwards (www.thl.fi/en/web/thlfi-en). Validity of the different diseases in the FHDR has been evaluated as satisfactory to good in numerous studies. The quality assessment of FHDR concerning the present cohort was performed prior to initiating this study (Saavalainen et al., 2018).

The Finnish Population Information System, maintained by the Finnish Population Register Centre, is a computerized national register that contains basic information on permanent residents, such as date of birth and death, address, and biological children of permanent residents (www.vrk.fi/en/frontpage).

Statistics Finland is a public authority that collects and maintains administrative data, such as the population census and the cause-of-death register (www.stat.fi/index_en.html). The latter includes the date of death and underlying cause of death according to the disease or circumstance (accident and act of violence) leading to death. The cause of death is determined according to the rules of the ICD-10 compiled by the World Health Organization. The causes of death are given by the treating physician and controlled regionally and at Statistics Finland.

Cohort Characteristics

The demographic characteristics of the whole study population (n = 148 780) were obtained from Statistics Finland. The population census data contain data on socioeconomic status, education, and profession. Because the data on occupation and socioeconomic status were limited to the census years (1995, 2000, and 2004–2012) only, we used the highest educational level from the 2014 census as a proxy for socioeconomic status. In the analysis, the highest education level reached was treated as a categorical variable with four categories: academic degree (bachelor, master, and doctoral), tertiary (short-cycle tertiary), upper secondary, and primary (primary and unknown).

The baseline calendar time, removal of gynecological organs, and parity status were represented as dichotomous variables. The baseline calendar time was divided into two periods according to the ICD coding system (ICD-9 in 1987–1995 and ICD-10 in 1996–2012). Based on the data on removal of the gynecological organs from the FHDR (1983–2012), we identified those with any gynecological organ removed before or at the index procedure. The baseline parity status was defined according to the information on live births obtained from the Population Register as at least one live birth before or at the index date and was then updated according to the follow-up information.

Follow-up and Outcomes of Interest

Women were followed up from the index day until death or until the end of follow-up on 31 December 2014, whichever came first. The outcome of interest was the mortality from any cause, as well as the cause-specific mortality based on the underlying cause of death. The specific causes of death were studied in groups formed according to the 54-group short-list of causes of death by Statistics Finland (Supplementary Table SI), where the alcohol-related deaths are separately presented as their incidence in Finland is high.

Statistical Analysis

For each outcome of interest, the crude mortality rate was calculated as the number of deaths divided by person time at risk, and the exact 95% confidence intervals (CIs) were assessed based on Poisson rates. For any cause mortality, we assessed both crude absolute and relative rate differences (ratio) in mortality rate between the cohorts. To allow for assessment of time-varying covariates, such as age, time since the index date, and parity, the individual follow-up time was split into the smaller bands. We calculated adjusted all-cause and cause-specific mortality rate ratios (MRRs) with 95% CIs by using multivariate Poisson model. We controlled for age and time since the index day as modeled by spline functions, the highest educational level and baseline calendar time (Model 1), and further for parity as assessed in a time-dependent manner (Model 2). The area of residence at the time of first endometriosis surgery was not statistically significant and therefore was not included. To study the changes in the all-cause MRRs during the follow-up, we plotted the adjusted MRR with 95% CI from the Model 2 along the time since the index date. In addition, we used the (multivariate) Poisson model with identity link to calculate the crude (adjusted) absolute rate differences with their 95% CIs for death from all-causes and major specific causes.

We performed several sensitivity analyses. To check whether the results were similar across all ages at death we calculated and plotted age-specific MRRs with 95% CIs for the all-cause mortality by six age groups (<30, 30–39, 40–49, 50–59, 60–69, and ≥70 years at death). To assess the potential heterogeneity in the MRRs according to the baseline gynecological organ removal in women with surgically verified endometriosis, we divided the endometriosis cohort into two groups, those with and without the baseline gynecological organ removal. By substituting the binary variable for endometriosis (no/yes) in Model 2 by a variable with three categories (no/yes without removal/yes with removal), we compared the endometriosis subgroups to the reference cohort, from which we excluded women who had undergone gynecological organ removal before beginning of the follow-up.

We set statistical significance level at 5% and considered the results with P<0.05 as statistically significant. All statistical analyses were performed using R statistical software version 3.5.0 (www.r-project.org), with the Epi package for splitting the individual follow-up (Plummer and Carstensen, 2011) and the Forestplot package for the graphical output (Gordon and Lumney, 2017).

Ethical Approval

Before initiation this study was approved by the ethics committee of the Hospital District of Helsinki and Uusimaa (238/13/03/03/2013). Permissions to utilize the data and to perform the linkages were provided by the National Institute for Health and Welfare (THL/546/5.05.00.2014), the Population Register Centre (D1794/410/14), and Statistics Finland (Dnro TK53-547-14).

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....