Comparing Swedish hospital discharge records with death certificates

Statistics Sweden. Department of Public Health and Caring Sciences, Unit of Social Medicine, Uppsala University, Sweden.
International Journal of Epidemiology (Impact Factor: 9.18). 07/2000; 29(3):495-502. DOI: 10.1093/ije/29.3.495
Source: PubMed


The quality of mortality statistics is of crucial importance to epidemiological research. Traditional editing techniques used by statistical offices capture only obvious errors in death certification. In this study we match Swedish hospital discharge data to death certificates and discuss the implications for mortality statistics.
Swedish death certificates for 1995 were linked to the national hospital discharge register. The resulting database comprised 69 818 individuals (75% of all deaths), 39 872 (43%) of whom died in hospital. The diagnostic statements were compared at Basic Tabulation List level.
The last main diagnosis and the underlying cause of death agreed in 46% of cases. Agreement decreased rapidly after discharge. For hospital deaths, the main diagnosis was reported on 83% of the certificates, but only on 46% of certificates for non-hospital deaths. Malignant neoplasms and other dramatic conditions showed the best agreement and were often reported as underlying causes. Conditions that might follow from some other disease were often reported as contributory causes, while symptomatic and some chronic conditions were often omitted. In 13% of cases, an ill-defined main condition was replaced by a more specific cause of death.
There is no apparent reason to question the death certificate if the main diagnosis and underlying cause agree, or if the main diagnosis is a probable complication of the stated underlying cause. However, cases in which the main diagnosis cannot be considered a complication of the reported underlying cause should be investigated, and assessments made of the feasibility and cost-effectiveness of routinely linking hospital records to death certificates.

38 Reads
  • Source
    • "The distribution of consistency and independence according to socio-demographic and medico-administrative variables gave expectable results: independence was more frequent in elder patients, likely because they suffer from multi-pathologies, or as the discharge-death time interval grew (8.5% of in-hospital deaths, 14.3% when death occurred within one month after discharge and 27.7% within 6 to 12 months), or for non-neoplasms UCDs, which had already been noted in former studies [5,6,8]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Background In the age of big data in healthcare, automated comparison of medical diagnoses in large scale databases is a key issue. Our objectives were: 1) to formally define and identify cases of independence between last hospitalization main diagnosis (MD) and death registry underlying cause of death (UCD) for deceased subjects hospitalized in their last year of life; 2) to study their distribution according to socio-demographic and medico-administrative variables; 3) to discuss the interest of this method in the specific context of hospital quality of care assessment. Methods 1) Elaboration of an algorithm comparing MD and UCD, relying on Iris, a coding system based on international standards. 2) Application to 421,460 beneficiaries of the general health insurance regime (which covers 70% of French population) hospitalized and deceased in 2008–2009. Results 1) Independence, was defined as MD and UCD belonging to different trains of events leading to death 2) Among the deaths analyzed automatically (91.7%), 8.5% of in-hospital deaths and 19.5% of out-of-hospital deaths were classified as independent. Independence was more frequent in elder patients, as well as when the discharge-death time interval grew (14.3% when death occurred within 30 days after discharge and 27.7% within 6 to 12 months) and for UCDs other than neoplasms. Conclusion Our algorithm can identify cases where death can be considered independent from the pathology treated in hospital. Excluding these deaths from the ones allocated to the hospitalization process could contribute to improve post-hospital mortality indicators. More generally, this method has the potential of being developed and used for other diagnoses comparisons across time periods or databases.
    BMC Medical Informatics and Decision Making 06/2014; 14(1):44. DOI:10.1186/1472-6947-14-44 · 1.83 Impact Factor
  • Source
    • "• Deaths from the Causes of Death Register (The National Board of Health and Welfare) which comprises all deaths since 1961 and is updated every year. The register covers all Swedish residents, whether the person in question was a Swedish citizen or not, and irrespective of whether the death occurred in Sweden or not [24] [25]. • The Prescribed Drug Register (The National Board of Health and Welfare) contains complete data on all prescription drugs dispensed in Sweden from July 2005 onwards, their amounts and dosages, as well as expenditures and reimbursement, and the age and gender of the patient regardless of reimbursement status, co-payment and prescriber category [26]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Earlier validation studies of risk scoring by CHA2DS2VASc for assessments of appropriateness of warfarin treatment in patients with atrial fibrillation have been performed solely with diagnoses recorded in hospital based care, even though many patients to a large extent are managed in primary care. Cross-sectional registry study of all 43 353 patients with a diagnosis of non-valvular atrial fibrillation recorded in inpatient care, specialist ambulatory care or primary care in the Stockholm County during 2006-2010. The mean CHA2DS2VASc score was 3.82 (4.67 for women and 3.14 for men). 64% of the entire cohort of patients with atrial fibrillation had the diagnosis in primary care (12% only there). The mean CHA2DS2VASc score of patients with a diagnosis only in inpatient care or specialist ambulatory care increased from 3.63 to 3.83 when comorbidities registered in primary care were added. In 2010 warfarin prescriptions were claimed by 47.2%, and ASA by 41.6% of the entire cohort. 34% of patients with CHA2DS2VASc=1 and 20% with CHA2DS2VASc=0 had warfarin treatment. ASA was more frequently used instead of warfarin among women and elderly patients. Registry CHA2DS2VASc scores were underestimated without co-morbidity data from primary care. Many individuals with scores 0 and 1 were treated with warfarin, despite poor documentation of clinical benefit. In contrast, warfarin appears to be underused and ASA overused among high risk atrial fibrillation patients. Lack of diagnoses from primary care underestimated CHA2DS2VASc scores and may thereby have overestimated treatment benefits in low-risk patients in earlier studies.
    International journal of cardiology 10/2013; 170(2). DOI:10.1016/j.ijcard.2013.10.063 · 4.04 Impact Factor
  • Source
    • "International Classification of Diseases (ICD) version 9 and subsequent versions provide information on metastatic sites as multiple causes of death, available at the Swedish Causes of Death Register. These data are highly reliable [12], partially because of the high proportion of deaths take place at hospitals [13], but also due to the fact that almost 95% of death certificates in cancer patients are based on examination at hospital prior to death [14]. We use here this source in combination with the TNM data to define survival in common cancers based on the site of metastasis. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Background Cancer of unknown primary site (CUP) is considered an aggressive metastatic disease but whether the prognosis differs from metastatic cancers of known primary site is not known. Such data may give insight into the biology of CUP and the metastatic process in general. Methods 6,745 cancer patients, with primary metastatic cancer at diagnosis, were identified from the Swedish Cancer Registry, and were compared with 2,881 patients with CUP. Patients were diagnosed and died between 2002 and 2008. The influence of the primary site, known or unknown, on survival in patients with metastases at specific locations was investigated. Hazard ratios (HRs) of death were estimated for several sites of metastasis, where patients with known primary sites were compared with CUP patients. Results Overall, patients with metastatic cancers with known primary sites had decreased hazards of death compared to CUP patients (HR = 0.69 [95% CI = 0.66–0.72]). The exceptions were cancer of the pancreas (1.71 [1.54–1.90]), liver (1.58 [1.36–1.85]), and stomach (1.16 [1.02–1.31]). For individual metastatic sites, patients with liver or bone metastases of known origin had better survival than those with CUP of the liver and bone. Patients with liver metastases of pancreatic origin had an increased risk of death compared with patients with CUP of the liver (1.25 [1.06–1.46]). The median survival time of CUP patients was three months. Conclusions Patients with CUP have poorer survival than patients with known primaries, except those with brain and respiratory system metastases. Of CUP sites, liver metastases had the worst prognosis. Survival in CUP was comparable to that in metastatic lung cancer. The aggressive behavior of CUP may be due to initial immunosuppression and immunoediting which may allow accumulation of mutations. Upon escape from the suppressed state an unstoppable tumor spread ensues. These novel data on the epidemiology of the metastatic process at the population level demonstrated large survival differences in organ defined metastases depending on the original cancer.
    BMC Cancer 01/2013; 13(1):36. DOI:10.1186/1471-2407-13-36 · 3.36 Impact Factor
Show more


38 Reads