methods
Multivariable Cox regression,
propensity score model
Multivariable Cox regression,
parametric
accelerated failure time models
Multivariable Cox regression
Multivariable Cox regression
DATA SOURCES
Study I, III and IV are based entirely on data from Swedish national healthcare registries.
Study II also uses local hospital registries, and on medical records. Linkage between registries, and between registries and medical records, was possible by using the Swedish personal 10-digit identity number, which uniquely identifies all Swedish residents.82
The Swedish Patient Registry
The Patient Registry was established in 1964, and contains complete nationwide data of in-hospital care in Sweden since 1987.83 Since 2001, the registry also contains complete data of specialist outpatient care and day surgery, including both private and public caregivers.
Diagnosis codes at discharge, codes of surgical procedures and hospitalisation dates are among the data that can be obtained. Validation studies have shown that 85-95% of diagnoses are valid,84 and codes representing upper gastrointestinal surgery have been shown to have up to 99.6% positive predictive value.85 The diagnoses are coded according to the International Classification of Diseases (ICD) versions 9 and 10.
The Swedish Cancer Registry
The Cancer Registry was established in 1958. It has 98% complete registration of type and date of gastric cancer diagnoses in Sweden according to a validation study from our group.86
The Swedish Causes of Death Registry
The Causes of Death Registry was established in 1961 in its current shape. It collects information about all deaths among Swedish residents and is believed to have a 99%
coverage.87
The Swedish Prescribed Drug Registry
The Prescribed Drug Registry records all prescribed and dispensed drugs in Sweden since 1st July 2005.88 The registry contains information on names of prescribed drug substances according to the anatomical therapeutic chemical classification (ATC).88 It also contains information about dose and amount of each prescribed drug.
The Swedish Registry of the Total Population
This Registry of the Total Population was established in 1968 and provides complete information on dates of birth, death, and migration in Sweden.
STUDY DESIGN AND METHODS
All studies included in this thesis were population-based cohort studies in design.
Study I
To compare mortality after less and more extensive surgery for peptic ulcer bleeding, Study I used data from the Swedish Patient Registry to identify all adult patients undergoing surgery for peptic ulcer bleeding between 1987 and 2008. We used the International Classification of Diseases (ICD) version 9 and 10 codes for peptic ulcer bleeding, and the Nordic Medico-Statistical Committee (NOMESCO) codes for identifying the relevant surgical procedures. Less extensive surgery was defined as under-running of the ulcer through a gastrotomy or duodenotomy with or without ligation of the major source artery and/or local excision of the ulcer. More extensive surgery was defined as resection of a part of the stomach or duodenum, with or without vagotomy. The Swedish Patient Registry was also used to identify comorbidities, and to identify if the hospital was a high volume, medium volume, or low volume centre for these procedures. Calendar period was taken into account by dividing the cohort into an early and a late period, analysed separately.
Outcomes were cause overall mortality, and 30-day, 90-day, 1-year, and 5-year all-cause mortality.
Study II
To compare key outcomes following transcatheter arterial embolisation (TAE) with conventional surgery for uncontrolled peptic ulcer bleeding, study IV identified patients undergoing TAE or surgery for peptic ulcer bleeding in Stockholm County between the years 2000 and 2014. Patients undergoing TAE were further evaluated through local hospital registries at the radiology departments. This procedure does not have an
established code in the Patient Registry. The registries used were the administrative sources on which the radiology departments get economic compensation for their examinations and interventions. All the departments used digitalised recording and patients have to be
registered in the system with their personal identity number in order to initiate an
examination or intervention. These individuals were then linked to the Patient Registry to identify those who had a peptic ulcer diagnosis at the same hospitalisation. All medical records for patients having undergone abdominal angiography at the time of hospitalisation for peptic ulcer were scrutinised by the author of this thesis, and patients with other
indications for angiography than peptic ulcer bleeding were excluded. Patients undergoing
surgery were identified through the Patient Registry. In patients who underwent both TAE and surgery, the first intervention after endoscopy was assigned to the individual. The primary outcome was all-cause mortality, occurring within 30 days, 90 days, 1 year, and 5 years after the intervention. Secondary outcomes were in-hospital bleeding,
re-intervention, duration of hospitalisation, and complications.
Study III
To assess risk factors for marginal ulcer after gastric bypass surgery, study II used the Patient Registry to identify all adult patients who underwent gastric bypass in Sweden between 2006 and 2011. The Patient Registry and the Prescribed Drug Registry were then used to identify the presence of any of 9 potential risk factors that were under study:
diabetes, hyperlipidaemia, hypertension, chronic obstructive pulmonary disease, ulcer history, and use of proton pump inhibitors (PPIs), aspirin, non-steroidal anti-inflammatory drugs (NSAIDs), and selective serotonin re-uptake inhibitors (SSRIs). The study outcome was the development of marginal ulcer.
Study IV
To test how various lengths of delays in H. pylori eradication influence the risk of recurrent peptic ulcer, ulcer complications (bleeding or perforation) and gastric cancer, Study IV used the Patient Registry and the Prescribed Drugs Registry to identify all adults in Sweden that were diagnosed with peptic ulcer and who were prescribed eradication therapy for H.
pylori between the years 2005 and 2013. Pre-defined time latency intervals between peptic ulcer diagnosis and H. pylori eradication were analysed in relation to the study outcomes.
STATISTICAL ANALYSES
Associations between exposures and outcomes in all studies were estimated using multivariable Cox proportional hazard regression models, which provided hazard ratios (HRs) and 95% confidence intervals (CIs), adjusted for pre-selected potential confounders. In Study I, an additional propensity score matched analysis was conducted since the hazards assumptions were not entirely met. In study II additional parametric accelerated failure time models were used to estimate the association between TAE or surgery and the outcome length of hospital stay, expressed as acceleration factor (AF) and 95% CI.
ETHICAL CONSIDERATIONS
All studies in this thesis were approved by the Regional Ethical Review Board in Stockholm. Study I, III and IV are strictly register-based, with study subject being
anonymous to the researcher. Study II included manual review of medical records, but data were analysed and presented at a group level. Data storage, management and analyses have been performed on firewall- and password protected servers at Karolinska Institutet. Discs were stored in locked safes located in constantly locked offices, accessed only by a
personal key card with password.