Smoking-attributable morbidity: acute care hospital diagnoses and days of treatment in Canada, 2002

Background Smoking is one of the most important risk factors for burden of disease. Our objective was to estimate the number of hospital diagnoses and days of treatment attributable to smoking for Canada, 2002. Methods Distribution of exposure was taken from a major national survey of Canada, the Canadian Community Health Survey. For chronic diseases, risk relations were taken from the published literature and combined with exposure to calculate age- and sex-specific smoking-attributable fractions (SAFs). For fire deaths, SAFs were taken directly from available statistics. Information on morbidity, with cause of illness coded according to the International Classification of Diseases version 10, was obtained from the Canadian Institute for Health Information. Results For Canada in 2002, 339,179 of all hospital diagnoses were estimated to be attributable to smoking and 2,210,155 acute care hospital days. Ischaemic heart disease was the largest single category in terms of hospital days accounting for 21 percent, followed by lung cancer at 9 percent. Smoking-attributable acute care hospital days cost over $2.5 billion in Canada in 2002. Conclusion Since the last major project produced estimates of this type, the rate of hospital days per 100,000 population has decreased by 33.8 percent. Several possible factors may have contributed to the decline in the rate of smoking-attributable hospital days: a drop in smoking prevalence, a decline in overall hospital days, and a shift in distribution of disease categories. Smoking remains a significant health, social, and economic burden in Canada.


Background
Smoking is responsible for high levels of morbidity and mortality. Smoking causes substantially increased risk of lung cancer, upper aerodigestive cancer, several other cancers, heart disease, stroke, chronic respiratory disease and a range of other medical conditions [1]. In 1996, a major study estimating the economic costs attributable to substance abuse in Canada (1992 data), was published [2]. As part of the analysis, the numbers of hospital separations and days of hospital treatment that could be attributed to smoking tobacco were estimated to be 208,095 and 3,024,265 respectively. This paper updates the estimates for the year 2002, using age-and sex-specific relative risks from recent meta-analyses and morbidity classified according to the International Classification of Disease version 10 (ICD-10).

Methods
The aim of the present study was to estimate the proportion of acute care hospital diagnoses and hospital days attributable to smoking in Canada for the year 2002.

Identification of diseases and meta-analyses on smoking risk relations
To identify the malignant and non-malignant health conditions for inclusion in the estimate, this analysis was guided by the 2004 Health Consequences of Smoking: A Report from the Surgeon General[1] which considers the following criteria in judgments of causality: consistency, strength of association, specificity, temporality, coherence, dose-response, and experimental evidence. The 2004 Surgeon General's report implemented a standardized, hierarchical language to summarize conclusions about causality, the highest of which is, "evidence is sufficient to infer causality". Only health outcomes for which the previous conclusion was reached for active smoking were included in the present analysis. Additionally, two conditions for which the causal role of passive smoking is conclusive were included.
Once identified, conditions were translated into corresponding ICD-10 codes. Finally, a comprehensive search strategy of current meta-analyses was performed for each disease category and its risk relationship with smoking.
Meta-analyses were identified using the Pubmed and OVID (1966 -January week 3,2005) databases. Search criteria were: smoking or tobacco, meta-analysis, and each disease category described in this paper.
Meta-analyses that included measures of smoking dose were preferred over those that only used current/former/ never categories. However, if relative risks (RRs) for doseresponse specific exposures were not found among the meta-analyses, current/former/never or ever/never categories were used. In 16 out of 20 disease categories included for active smokers, dose-response specific RRs were available. In two categories, plus one for smokers aged 65 or older, only current/former/never RRs were available. In two categories, only ever/never RRs were available. Similarly, analyses that included age-and sex-stratified estimates of relative risk were preferred over more crude estimates. Age-specific estimates were available for ischaemic heart disease (IHD), cardiac arrhythmias, heart failure, and cerebrovascular disease. In cases where a more recently published meta-analysis did not exist, the relative risk from English et al [3] was used. When a meta-analysis was published later than 1995, there was usually only one that presented data on smoking dose, so it was used as the source of relative risk. If there was more than one, all were examined, and that of highest quality and the most comprehensive based on age and smoking dose categories was chosen.
The ICD-10 codes of all disease categories related to smoking included in the present analysis are presented in Table 1.
Passive smoking-attributable morbidity was derived by applying sex-specific RRs and rates of morbidity from lung cancer and age-and sex-specific RRs and rates of morbidity from IHD to the population of Canadians who have never smoked but are exposed to environmental tobacco smoke (ETS) in the home from spouses and other sources. RR estimates were obtained from the most comprehensive meta-analyses applicable to Canada. This procedure did not include the effect of ETS exposure on current smokers and thus the total burden of smoking was underestimated.
All RRs used may be obtained from the comprehensive report of this cost study [4].

Prevalence of smoking in Canada
Smoking prevalence for different levels of smoking consumption for Canada as a whole were obtained from the Canadian Community Health Survey 2003 (CCHS cycle 2.1) [5], a population based representative survey conducted by Statistics Canada. All prevalence estimates were sex-and age group-specific. The categorizations of smoking status, however, varied by disease based on the RRs available in the meta-analyses; for example, for COPD, RRs risks were available for current, former and never smokers, thus the prevalence of current, former and never smokers was used. For each disease for which the identified meta-analysis included dose-response-specific RRs, prevalence estimates also were dose-specific (e.g., never, former, current, 1-14, 15-24, 25+ cigarettes per day). Current smokers, those who reported occasional smoking or daily smoking, were further categorized by number of cigarettes per day when sufficient information existed to do so.
The prevalence of non-smokers living inside a home where another person smokes was also available from the CCHS data set and was used to calculate 2002 passive smoking hospital diagnoses and days.
The sample weights provided by Statistics Canada to ensure comparability between the CCHS sample and the Canadian population were used to calculate prevalence based on sex and age groups. The age groups used were 15-29, 30-44, 45-59, 60-69, 70-79, and 80+.

Morbidity data
Acute care hospital diagnoses and days data in Canada for 2002 were obtained from the Canadian Institute for Health Information (CIHI) on national and provincial levels according to ICD-10. For the national level, data were provided for each disease condition as well as for each sex and five-year age group from 0 to 80. However, the national level data were actually composed of only seven provinces and two territories (Alberta, British Columbia, Newfoundland, Northwest Territories, Nova Scotia, Ontario, Prince Edward Island, Saskatchewan, Yukon Territory). Based on these, data for Canada as a whole were estimated using the total population: the disease-specific rate of occurrence observed in the data provided was applied to the total population of Canada to obtain the estimated number of disease-specific occurrences.
The Hospital Morbidity Database (HMDB) held by CIHI captures information on patients separated (through discharge or death) from acute care facilities in Canada. As such, it provides national data on acute care hospitalizations by diagnoses and procedures. Day procedures (e.g., day surgeries), outpatient, and emergency department visits are not captured in this database. Stillborns and cadaveric donor "discharges" are excluded, whereas newborns are included in the HMBD database. Figures are based on facility geography, that is, where the hospital is located (i.e., it may include non-Canadians). Additionally, the statistics reflect the number of hospitalizations, which is somewhat higher than the number of individuals diagnosed since individuals with multiple admissions during a single year would be counted more than once.
Hospital days (Length of Stay) is associated with the condition coded as Most Responsible Diagnosis (MRD) on the patient's hospital record. In other words, the MRD accounts for the majority of a patient's days in hospital. MRD is the one diagnosis that describes the most significant condition of the patient that is responsible for his/ her stay in hospital. In a case where multiple diagnoses may be classified as most responsible, coders are instructed to code the diagnosis responsible for the longest length of stay [6]. However, there is still overlap in the database, i.e. more than one diagnosis listed per hospital stay. Thus, to calculate hospital days attributable to tobacco smoking without overlap, the disease-specific hospital days were scaled down by the factor of overall hospital days in Canada divided by the total of the disease-specific hospital days.

Computing smoking-attributable fractions
The smoking-attributable fraction (SAF) is defined as the fraction of the disease in the population that would not have occurred if the effect associated with smoking was absent [7,8]. SAFs were assessed for different specific causes of illnesses by two methods: • chronic disease SAFs were calculated by combining exposure from CCHS and RR estimates from meta-analyses, • fire injury was calculated using direct estimates of smoking involvement from the Council of Canadian Fire Marshals and Fire Commissioners [9].
We used the most comprehensive meta-analysis for each condition, as described above (see also Table 1). The RR for each condition was combined with different levels of smoking consumption for each sex and age group and an attributable fraction was obtained using the following formula (see Walter, 1976Walter, , 1980.
i: exposure category with baseline exposure or no smoking i = 0.
RR i : relative risk at exposure level i compared to no consumption.
P i : prevalence of the ith category of exposure.
The SAFs were then applied to the hospital diagnosis and length of stay information to estimate the smoking-attributable morbidity by age and sex. To illustrate this procedure, a simple example follows. For chronic obstructive pulmonary disorder (COPD) the all-age and all-sex RRs for current smokers, and former smokers, relative to never smokers are 9.80, 6.70. The prevalence, with rounding error, of current, former, and never smokers among men aged 45-59 were 26.0, 53.0, and 21.1. Simple algebra produces a SAF of 0.84, which is applied to the 978 hospital diagnoses of COPD among all Canadian men of this age group to determine that 823 of the diagnoses were attributable to smoking (355 and 468 among the current and former smokers respectively). It is notable that although the RR among current smokers is greater than that among former smokers, the greater prevalence of former smokers among men aged 45-59 resulted in a larger smokingattributable burden of COPD among the former smokers.

Estimating costs of acute care hospital days
Daily average per-capita cost for acute care hospital separations for each available province and territory was obtained from CIHI [10]. For Nunavut, average per-capita cost was not available; therefore, average per-capita cost for Northwest Territories was substituted. To obtain the total costs for each province and territory, daily average per-capita cost was applied to the total number of smoking-attributable acute care hospital days. These numbers were summed to obtain the total for Canada. A similar procedure was followed for those acute care hospital days attributable to alcohol and illegal drugs for comparison purposes. Table 2 gives an overview of the estimated smoking exposure in Canada by sex and age group. As expected, men smoked more than women on average, and smoking prevalence decreased with age.

Conclusion
Overall, there were 2,210,155 acute care hospital days attributable to smoking, accounting for 10.3 percent of all those in Canada in 2002. At the time of the last cost study (data for 1992) for which similar estimates were produced, the number of acute care hospital days attributed to smoking was higher at 3,024,265, which accounted for 7.3 percent of all acute care hospital days. In considering these numbers, it must be noted that the total number of hospital days due to overall morbidity has decreased due to changes in the Canadian health care system meant to reduce health care expenditures such as progress in treat-  Although the same approach was utilized in calculating the smoking-attributable fractions in both 1992 and 2002, different inputs were used. That is, the relative risks that were used to estimate the relationships between smoking and specific health outcomes were updated to reflect recently available literature. Also, this analysis, in determining which health outcomes to include in these estimates, was guided by the report of the Surgeon General and thus included acute myeloid leukaemia but did not include chemotherapy under a separate heading. This may be suspected of causing significant differences in the estimates for the two years, however in one of a number of sensitivity analyses published in a recent article, when 2002 data were analyzed using the 1992 approach in addition to our present approach the estimates were very similar (2,316,166 versus 2,210,155) [14].
The smoking-attributable hospital diagnoses estimates produced in this analysis cannot be compared with the first Canadian cost study, as that work measured hospital separations and not diagnoses.
The burden of smoking-attributable morbidity as measured by hospital diagnoses has decreased substantially since the last Canadian cost study, which estimated the burden of morbidity attributable to smoking in Canada. While this is reason for optimism, smoking continues to present a serious health concern in Canada; smoking causes substantial morbidity in this country, and carries huge humanitarian and financial costs. In 2002, smokingattributable acute care hospital days alone cost Canadian taxpayers in excess of 2.5 billion dollars [4], far more than alcohol and illegal drug use consumption combined. Given concerns about the increasing cost of the Canadian health care system and the pressures to preserve this same system, here is a viable target for cost savings. Smokingattributable morbidity is avoidable. Thus, it is vital that policy and policy-makers address this substantial health, social, and economic burden in Canada.
helped to draft the manuscript. All authors read and approved the final manuscript.