Surveillance of US Death Rates from Chronic Diseases Related to Excessive Alcohol Use

Transcription

1 Alcohol and Alcoholism, 2016, 51(1) doi: /alcalc/agv056 Advance Access Publication Date: 3 June 2015 Article Article Surveillance of US Death Rates from Chronic Diseases Related to Excessive Alcohol Use Anthony P. Polednak* Retired, Connecticut Department of Public Health, 410 Capitol Avenue, Hartford, CT 06134, USA *Corresponding author: appoled7@yahoo.com Received 9 February 2015; Revised 5 May 2015; Accepted 9 May 2015 Abstract Aims: To assess the utility of multiple-cause (MC) death records for surveillance of US mortality rates from chronic causes related to excessive alcohol use. Methods: The Alcohol-Related Disease Impact (ARDI) resource produced estimates of the population alcohol attributable fraction (AAF) due to excessive drinking for each alcohol-related (AAF > 0%) cause of death, and used AAFs to estimate numbers of alcohol-related deaths from alcohol-related underlying causes (UC) in adults age and 65+ years in For surveillance, this study used MC death file to identify individual deaths ( ) with an alcohol-induced cause (AAF = 100%) anywhere on the certificate, and to obtain US rates of premature death (ages and years) for Results: Using the selected MC records, numbers of deaths from alcohol-related chronic UC ( ) were 81% of ARDI estimates for age 20 64, but only 40% for 65+ years. The MC records identified substantial numbers of deaths from causes (e.g. certain infectious diseases) not included as alcohol-related in ARDI, but included in surveillance of premature death rates for chronic UC. Also, premature death rates for chronic alcohol-induced causes using only the UC (as in routine mortality statistics) were only about half the rates based on MC; all rates increased in recent years but some reached statistical significance only by using MC. Conclusions: Using MC records underestimated total US deaths from alcohol-related chronic causes as the UC, but enhanced surveillance of rates for premature deaths involving chronic causes that may be related to excessive alcohol use. INTRODUCTION The population alcohol attributable fraction (AAF) may be interpreted as the estimated proportion of all cases or deaths from a particular condition in a population that may be due to alcohol, assuming a causal association. AAFs are estimated statistically by formulas that include the estimated relative risk for each condition or cause of death according to specific levels of drinking, and the prevalence (proportion) of drinkers at each of these levels in population. These calculated AAFs are applied to mortality data, in order to estimate numbers of deaths from conditions either partially or wholly attributable to alcohol. Estimated AAFs and estimated alcohol-related deaths have been reported globally (Rehm and Shield, 2014), and for individual countries such as Scotland (Information Services Division, 2009), Ireland (Martin et al., 2010) and Denmark (Eliasen et al., 2014). Estimated numbers of alcohol-related deaths, and their proportion of all deaths in the population, were several times as large as numbers from routine surveillance limited to causes of death wholly attributable to alcohol in Scotland (Information Services Division, 2009) and Ireland (Martin et al., 2010). Such findings on the potential burden of alcohol-related deaths have important implications for health policy and planning. Data based on estimated AAFs in a population can be used to assess the results of efforts to enhance the utility of death certificates for surveillance of alcohol-related mortality, as done in the present study using US databases. The US Centers for Disease Control and Prevention s (CDC s) Alcohol-Related Disease Impact (ARDI) resource has estimated the AAF for each of 54 causes of death. ARDI uses these AAFs to estimate average annual numbers of deaths attributable to The Author Medical Council on Alcohol and Oxford University Press. All rights reserved 54

2 Alcohol and Alcoholism, 2016, Vol. 51, No excessive alcohol use or (alternatively) any alcohol use in (CDC, 2013). Certifiers completing death records may be especially aware of a decedent s condition(s) that is wholly attributable to excessive alcohol use, which may be common among hospital inpatients and outpatients (Saitz et al., 2006; Johnson et al., 2014). ARDI, however, uses only the underlying cause of death from the death certificate, and does not consider all causes mentioned ( multiple causes or mentions ). The first aim of this study was to use multiple-cause records that mentioned an alcohol-induced cause (AAF = 100%, or wholly attributable to alcohol) anywhere on the certificate, as an indicator of the probable role of excessive alcohol use in an individual death (see Methods section for details). The annual numbers of these deaths in from alcohol-related underlying causes were compared with ARDI estimates (as a gold standard) which were available for (CDC, 2013). ARDI is not intended to be used to monitor changes over time in numbers of estimated deaths (CDC, 2013; Stahre et al., 2014), and does not provide estimates of death rates from alcohol-related causes. The second aim of this study was to use the numbers of alcohol-related deaths obtained from multiple-cause death files (available for ) to calculate alcohol-related mortality rates in the population, for surveillance of trends in alcohol-related death rates. These analyses also included a comparison of US death rates for alcohol-induced chronic causes based on multiple causes of death vs. rates based on the underlying cause alone (as in routine mortality statistics). In Northern Ireland, using multiple causes of death (including contributory causes) vs. the underlying cause alone from death records resulted in a 40% increase in the number of deaths for a group of causes with an AAF of 100% plus certain chronic liver diseases with AAF < 100%; death rates in the population were not reported (Durkin et al., 2010). The focus of this study was on chronic causes, because alcoholrelated acute or external causes (e.g. accidental injuries and suicides) are known to be underestimated from death records (Pollock et al., 1987; Moyer et al., 1989) even using multiple causes of death (Castle et al., 2014). METHODS Estimated deaths from alcohol-related causes due to excessive alcohol based on AAFs using the ARDI resource Estimated average annual numbers of deaths from excessive alcohol use for alcohol-related causes were obtained for from the ARDI resource; methods are described elsewhere (CDC, 2013). Briefly, for each chronic cause of death, ARDI calculates the AAF based on estimates of relative risks (at specific levelsof drinking) obtained from published meta-analyses of epidemiologic studies. For most chronic causes, excessive (i.e. medium or high ) use in average number of drinks (14 g per drink) per day is 2.9+ for males vs for females, whereas low use is for males vs for females. Calculating the AAF requires the estimated prevalence (i.e. proportion) of the US adult population age 20+ years by level of drinking. ARDI uses self-reported data on current drinking (i.e. in the past 30 days), including number of drinksandbingedrinkingepisodes, fromthe2010cdc s Behavioral Risk Factor Surveillance System s survey(cdc, 2013). For acute causes (such as injuries), estimated AAF are derived from studies using blood alcohol level at death or other estimates of alcohol use (CDC, 2013). The AAFs in ARDI are not age-specific but differ by sex for some causes (CDC, 2013). In ARDI, the AAF for each specific causeof death is multiplied by the total number of US deaths from that cause (as the underlying cause of death), by age group and sex, as provided by the National Center for Health Statistics (NCHS). This produces estimated annual numbers of alcohol-attributable deaths due to excessive alcohol use for each cause (CDC, 2013). For excessive alcohol use, the estimated numbers involve no deductions for putative protective effects of alcohol use on ischaemic heart disease (IHD) deaths (CDC, 2013; Stahre et al., 2014). The only deduction is 10 deaths from cholelithiasis (CDC, 2013) which were ignored in this study. For IHD, the estimated relative risks based on no current drinking as the reference category were all <1.00 i.e for low, 0.84 for medium and 0.88 for high average daily alcohol consumption (CDC, 2013). For IHD and other chronic causes with AAF < 100%, ARDI calculates relative risks by using low consumption as the reference category (CDC, 2013). This resulted in an estimated 738 deaths per year from IHD attributed to excessive alcohol use at age 20+ years in (Stahre et al., 2014). Using these 738 deaths, and the total US deaths from IHD at age 20+ years recorded in ARDI, the AAF due to excessive drinking would be (reported as 0.00 in CDC, 2013). IHD was not included among alcohol-related causes in the present study, however, because any protective effects of low vs. no current alcohol use (if real) may be limited to certain subgroups defined by gender and age, and to drinkers without pre-existing health conditions (Knott et al., 2015) or episodes of heavy drinking (Roerecke et al., 2011). Alcohol-related underlying causes of death, coded to the International Classification of Diseases Version 10 (ICD-10), are classified as either chronic or acute in ARDI (CDC, 2013). For this study, chronic causes were grouped as cancer (all with AAF < 100%), and mental and nervous, digestive, and circulatory (cardiovascular) systems. Acute or external causes almost exclusively involved AAFs of <100%. Details and ICD-10 codes for each group of causes are provided in Table 1. Not included in ARDI, however, are alcohol-related infectious causes (e.g. hepatitis C infection), due to inadequate data for estimating AAFs (CDC, 2013; Gonzales et al., 2014). Numbers of US deaths from alcohol-related causes using multiple-cause death files, compared with estimated deaths obtained from ARDI The National Center for Health Statistics (NCHS) Multiple Cause of Death Data File (CDC, 2014a) includes de-identified data from death records for all US residents ( ), with causes of death coded to ICD-10 (World Health Organization [WHO], 2011). Causes of death are based on automated coding systems that standardize the application of ICD-10 rules including selection of one underlying cause for each decedent. The database (CDC, 2014a) included, in addition to the underlying cause, as many as 20 causes for each decedent that are either complications of the underlying cause (on part I of the certificate) or other significant conditions contributing to death (on part II). Alcohol-induced (AAF = 100%) causes of death in the NCHS file were the same as in ARDI, except that NCHS included alcoholinduced acute pancreatitis (ICD-10 K85.2) whereas ARDI defined all acute pancreatitis (K85) as <100% AAF (CDC, 2013). Starting with deaths in 2007, ICD-10 coding rules classified acute intoxication (formerly code F10.0) to external causes. In addition, deaths with any

3 56 Alcohol and Alcoholism, 2016, Vol. 51, No. 1 Table 1. Average annual numbers of US deaths at ages years and 65+ years in from an alcohol-related underlying cause: using alcohol-related disease impact (ARDI) resource estimates for excessive alcohol use (CDC, 2013) vs. using multiple-cause death records mentioning any alcohol-induced condition (alcohol-attributable fraction = 100%) Underlying cause of death category Age Years Age 65+ Years ARDI Death ARDI Death Resource Records Resource Records No./year No./year No./year No./year 1. Alcohol-related chronic causes Cancers a Mental and nervous systems b Mental-behavioural disorders due to alcohol c (5047) (5051) (1356) (1354) Digestive system d 16,361 12, Unspecified liver cirrhosis (4281) (85) (3566) (16) Alcoholic liver disease c (11,501) (11,814) (2863) (2882) Circulatory system e Alcoholic cardiomyopathy c (373) (376) (140) (141) Hypertension (407) (823) (1058) (276) Stroke, ischaemic or haemorrhagic (666) (240) (1507) (77) Subtotal, all chronic alcohol-related (24,198) (19,579) (13,200) (5262) 2. Alcohol-related acute (i.e. external) causes f 36, , Alcohol poisoning; suicide by alcohol c (1552) (1551) (98) (99) 3. Total alcohol-related causes (as defined in ARDI) 60,612 29,253 23, Any underlying cause NA 37,425 g NA 9342 g ICD-10, International Classification of Diseases Version 10 (see text); NA, not applicable. a Causes with AAF = 100%: none. AAF<100% causes (and ICD 10 codes): oropharynx (C01 06, C09 10, C12 14), esophagus (C15), liver (C22), larynx (C32), breast (C20), prostate (C61). b AAF = 100%: alcohol abuse/dependence syndrome, alcoholic psychosis (F ), degeneration of nervous system due to alcohol (G31.2), alcoholic polyneuropathy (G62.1) alcoholic myopathy (G72.1). AAF < 100%: epilepsy (G40-41). c For these causes (AAF = 100%), numbers from ARDI should agree with those from death records used in this study (see text), except for small differences presumably due to rounding. For alcoholic liver disease, however, ARDI excluded deaths coded to K70.0 (alcoholic fatty liver), which were included in the number shown for death records. d AAF = 100%: alcoholic gastritis (K29.2), alcoholic liver disease (K70), alcohol-induced chronic pancreatitis (K86.0). Causes with AAF < 100%: gastro-oesophageal haemorrhage (K22.6), acute pancreatitis (K85), chronic pancreatitis (K86.1), chronic hepatitis (K73), unspecified liver cirrhosis (K , K76.0, K76.9), portal hypertension (K76.6), cholelithiasis (K80). e AAF = 100%: alcoholic cardiomyopathy (I42.6). AAF < 100%: hypertension (I10 I15), ischaemic and haemorrhagic stroke (G45, I63, I65 67, I69.3, I60 62, I ), oesophageal varices (I85 and I98.2), supraventricular cardiac dysrhythmia (I47.1, I47.9, I48). This study did not include ischaemic heart disease (I20 25) among the causes related to excessive alcohol use (see text). f AAF = 100%: excessive blood alcohol level (R78.0), and alcohol poisoning or alcohol exposure of accidental, intentional or undetermined intent (X45, X65, Y15). AAF < 100%: accidental injuries, homicide and suicide (i.e. V01 97, and selected W, X and Y codes) (CDC, 2013). g Includes deaths from causes not included in ARDI (CDC, 2013) (see text). F10 code that also mentioned certain liver diseases within codes K72-K76.9 on the same record were coded to alcoholic liver disease (K70) (WHO, 2011; CDC, 2014a). This resulted in an artefactual increase in numbers and rates for code K70, and a decline for code F10.0, from 2006 to The method of using all deaths with mention of an alcoholinduced cause anywhere on the death certificate identifies the subgroup of alcohol-related causes (AAF > 0%) that have an alcoholinduced cause (AAF = 100%) as the underlying cause. For each of these causes, numbers of deaths obtained should agree with those ARDI estimates, because ARDI simply multiplies by 100% the number of deaths from each of these causes as obtained from an NCHS mortality database (CDC, 2013). For identifying individual deaths from any alcohol-related cause with an AAF < 100% that involved excessive alcohol use, however, there is no item on the death certificate asking the certifier about the probable role of alcohol in contributing to death. Moreover, such a judgment for any individual decedent would be speculative, as also noted for all tobacco-related causes (Peto and Doll, 1992). Instead, this study identified individual multiple-cause death records that had an alcohol-related underlying cause with AAF < 100% and also mentioned (elsewhere on the same death certificate) any alcohol-induced cause (AAF = 100%) as definedinthenchsfile (CDC, 2014a). These deaths were regarded as probably involving excessive alcohol use. Most alcohol-induced conditions (e.g. alcohol abuse or dependence, alcoholic psychosis, alcoholic liver disease, and alcohol poisoning) are indicative of current and/or past excessive alcohol use. Comparisons of total numbers of alcohol-related deaths obtained in this study with ARDI estimates were made for age groups 20 34, and years combined (i.e years) and for 65+ years (i.e. the only older age category available from ARDI) (CDC, 2013). The study method also will identify deaths from an alcohol-related infectious disease (which are not included in ARDI), or any other underlying cause not defined as alcohol-related in ARDI. These additional deaths were included in some analyses of premature mortality rates.

4 Alcohol and Alcoholism, 2016, Vol. 51, No Analysis of death rates ( ) using multiple-cause death records The CDC database used the latest available population figures for the US resident population from the 2010 census and population estimates for intercensal years as denominators in calculating death rates (CDC, 2014a). For analyses of trends ( ) in US death rates from alcohol-related chronic causes, this study used annual age-standardized mortality rates (ASMRs) to adjust for changes in age composition of the US population. Direct standardization to the age distribution of the US 2000 standard population required including the age group years (CDC 2014a); hence, age years was used instead of years. The age-specific death rate for years was also analysed, because premature mortality in the US (defined historically as <65 years) is currently defined as <75 years (NCHS, 2014). The standard error (SE) of each rate was obtained and 95% the confidence interval (CI) was estimated as the rate±[2 SE] (CDC, 2014a). Data files with rates and their SEs were imported into the National Cancer Institute s Joinpoint Regression Program (version 4.0.1, January 2013). The program uses the natural logarithm of the rate, with calendar year as the independent variable, and least squares regression to identify changes in the linear slope over time and a permutations test that selects the best-fitting model (Kim et al., 2000). Annual percent change (APC) in the rate and the upper and lower 95% confidence limits (CL) of the CI were obtained for each segment, if more than one (e.g. one joinpoint involves two line segments). The maximum number of joinpoints was 2 (i.e. 3 line segments), based on program requirements (e.g. 4+ observations between two joinpoints, and 3 observations from a joinpoint to the end of the 14-year period). The average APC (AAPC) for the past 5 years, a weighted average of the APCs, was used to summarize recent trends in the same direction but differing in magnitude (Clegg et al., 2009). RESULTS Deaths from alcohol-related causes as underlying cause using ARDI vs. death records For age years, using the study method involving multiple-cause death records, identified 29,253 deaths/year with a mention of an alcohol-induced cause that had an alcohol-related underlying cause (Table 1). These 29,253 deaths were only 48.3% of the number (60,612) obtained from ARDI (Table 1). In this age group numbers obtained from death records were especially low for external causes (i.e. 9674/year or 26.6% of the 36,414/year from ARDI), but much higher for chronic causes (19,579/year or 80.9% of the 24,198 from ARDI) (Table 1). Of the 19,579 deaths from chronic causes, 11.1% were from causes with AAF < 100% (vs. 29.1% in ARDI) (not shown). For age years agreement in numbers was very close (98.0%) for alcohol-related mental and nervous system causes, all of which had an AAF of 100% except for epilepsy (AAF = 15% in ARDI) (CDC, 2013) (Table 1). For all chronic alcohol-related causes combined, the lower agreement was largely due to unspecified liver cirrhosis (AAF <100%) (Table 1). Agreement was close for circulatory system (cardiovascular) causes, despite the fact that only alcoholic cardiomyopathy (<400 deaths/year) has an AAF of 100%, but the study method resulted in about twice as many deaths from hypertension and only about a third from stroke, compared with ARDI estimates for age years (Table 1). For cancers (all with AAF <100%), the number using death record was 58.1% of the ARDI estimate. The 37,425 deaths/year from any underlying cause with mention of an alcohol-induced cause at age years (Table 1) included 8471 not defined as alcohol-related in this study. Of these 8741 deaths (not tabulated), 2915 were coded to cardiovascular conditions as the underlying cause, including 2137 from IHD (codes I20 I25) vs. 143 IHD deaths obtained by using the estimated AAF for IHD from ARDI (see Methods section). Other common underlying causes were not in ARDI, including 1436 deaths from infectious diseases (A00-B99), which were predominantly chronic hepatitis C (code B18.2, n = 983) and HIV disease (code B20, n = 98). In addition, 851 deaths were from respiratory system causes (codes J00-J98), 569 from endocrine, nutritional, metabolic (E00 E88), 544 from neoplasms (e.g. pancreas, lung, and colon and rectum), 446 from digestive system, 820 from external causes, and small numbers of various other chronic causes. For age 65+ years, death records yielded much lower numbers of deaths vs. ARDI, for most chronic causes as well as for acute causes. For chronic alcohol-related causes, the method using multiple-cause death records resulted in only 5262 alcohol-related chronic causes or 39.9% of the 13,200 from ARDI (Table 1). Of these 5262 deaths, 15.0% were from causes with AAF < 100% (vs. 62.7% of deaths in ARDI) (data not shown). Agreement was close only for mentalnervous system causes. Discrepancies in numbers of deaths vs. ARDI estimates were largely due to unspecified liver cirrhosis, along with hypertension and stroke (circulatory system) and cancers (Table 1). For age 65+ years, the total of 9,342 average annual deaths with mention of an alcohol-induced cause included 3434 from causes not defined as alcohol-related in ARDI (Table 1) were mainly cardiovascular (1404 or 40.9%), respiratory (644 or 18.8%), neoplasms (469 or 13.7%), endocrine-metabolic (222 or 6.5%) and infectious (178 or 5.2%). Cardiovascular causes included 1016/year from IHD; in comparison, the ARDI estimate based on AAF was 595. Temporal trend ( ) in death rates using multiple-cause records age years In view of under-ascertainment of chronic alcohol-related deaths at age 65+ years vs. ARDI (Table 1), analyses of trends using death records focused on age years. Using only the underlying cause of death, and selecting deaths from any alcohol-induced chronic cause (i.e. AAF = 100%, codes F , G31.2, G62.1, G72.1, I42.6, K29.2, K70 and K86.0) as the underlying cause, the annual ASMRs for age years were remarkably similar (i.e ) from 1999 through 2009, but increased to 8.5 in 2012 (Table 2). These ASMRs were lower than (and their CIs did not overlap with) ASMRs obtained by the method used in this study to identify alcohol-related deaths (Table 2, under Multiple Causes ). Even higher ASMRs were obtained by selecting deaths from any chronic underlying cause, including those not defined as alcohol-related in this study (Table 2). The recent increases in the rates from 2009 to 2012 (Table 2) were due mainly to alcoholic liver disease (i.e. from 5.49 to 5.99 per 100,000), liver cancer (C22) (i.e. from 0.18 to 0.26 per 100,000) as the underlying cause, along with a smaller increase for IHD. For each of these causes, the CIs on the rates for 2009 and 2012 did not overlap (data not tabulated). Trends in rates for premature death from alcoholinduced chronic causes (AAF = 100%) using multiple causes vs. the underlying cause alone: age <75 years Because only alcohol-induced causes as the underlying cause are reported in routine mortality statistics, an aim of this study was

5 58 Alcohol and Alcoholism, 2016, Vol. 51, No. 1 Table 2. Annual ( ) age-standardized death rate per 100,000 in US residents age years a using only the underlying cause (UC) of death and selecting any alcohol-induced b chronic cause, vs. using death records with mention of any alcohol-induced cause and selecting any chronic alcohol-related causes or any chronic cause as UC UC only Multiple causes (mention of any alcohol-induced cause) UC: chronic alcohol-induced c UC: chronic alcohol-related d UC: any chronic cause of death e Year No. Rate (CI) No. Rate (CI) No. Rate (CI) , ( ) 15, ( ) 20, ( ) , ( ) 15, ( ) 21, ( ) , ( ) 16, ( ) 21, ( ) , ( ) 16, ( ) 22, ( ) , ( ) 16, ( ) 23, ( ) , ( ) 17, ( ) 24, ( ) , ( ) 18, ( ) 25, ( ) , ( ) 18, ( ) 25, (11.8, 12.1) , ( ) 18, ( ) 26, ( ) , ( ) 19, ( ) 26, ( ) , ( ) 19, ( ) 27, ( ) , ( ) 20, ( ) 28, ( ) , ( ) 21, ( ) 29, ( ) , ( ) 22, ( ) 31, ( ) , ( ) 257, ( ) 354, ( ) Joinpoint regression analysis of rates: segments (years) and APCs Years APC, % (CI) APC, % (CI) APC, % (CI) ( 0.2, ±0.1) 0.5 ( )* 0.4 ( )* ( )* 3.4 ( )* 3.0 ( )* APC, annual percent change in rate, from joinpoint regression (see text); CI, confidence interval (95%), lower and upper limits; ICD-10, International Classification of Diseases Version 10 (see text). a Age years was used to calculate the age-standardized death rate (CDC, 2014a) (see text). b As defined in a US national mortality data resource (CDC, 2014a) (see text). c Any alcohol-induced chronic cause, with alcohol attributable fraction (AAF) of 100%; excludes ICD-10 code F10.0 (acute alcohol intoxication) (see text for explanation). d Any alcohol-related chronic cause (AAF > 0%) as defined in ARDI (CDC, 2013) except that ischaemic heart disease was not included in the study (see Table 1 and text). e Any underlying cause excluding external causes (see Table 1 and text). *P <.05 for APC. to compare trends in premature mortality rates (age <75 years) for alcohol-induced chronic causes (AAF = 100%) using the underlying cause alone vs. using multiple causes. Rates were examined only for (Table 3), or since the advent of changes in coding rules that affected numbers of deaths (starting in 2007) coded to alcoholic liver disease (code K70) and mentalbehavioural disorders due to alcohol (codes F ) (see Methods section). Using the underlying cause only, deaths coded to either K70 or mental-behavioural disorders due to alcohol (codes F ) together comprised >95% of all deaths from any alcohol-induced chronic cause at age years and also at years (Table 3). Recent increases in rates at age years were large and similar in magnitude whether based on multiple causes of death or the underlying cause alone. For rates at age years, however, the change in rate from 2007 to 2012 for code K70 was slightly larger using multiple causes. For age years, for codes F10.1 F10.9, the rates were >3 times as high based multiple causes (vs. underlying cause alone), yielding an APC that was statistically more reliable and reached statistical significance (Table 3). The recent increase was due to alcoholic liver disease but also cardiovascular causes (not alcohol-related) as underlying cause (data not tabulated). DISCUSSION Limitations of ARDI estimates of numbers of alcohol-related deaths For estimating deaths from excessive alcohol use, limitations of ARDI as the gold standard include inaccuracies in the prevalence of excessive current alcohol use from surveys and (hence) in the AAFs for specific causes of death (CDC, 2013; Shield et al., 2013, 2014). ARDI does not address former drinkers, uses only the underlying cause of death, and omits alcohol-related infectious causes (CDC, 2014b; Stahre et al., 2014) and colorectal cancer which is included in other estimates of alcohol-related cancers (Shield et al., 2013, 2014). The use of low-level drinking as the reference category in estimating AAFs due to excessive drinking for some chronic causes (CDC, 2013) probably underestimates alcohol-related deaths (except for IHD). With regard to estimating premature alcohol-related deaths (age <75 years), ARDI produces estimates only for age groups within years and for the broad group of 65+ years (CDC, 2013). Numbers of alcohol-related deaths for age years could be obtained by using the AAFs available in ARDI, but the lack of age-specific estimates of AAFs in ARDI is a potential issue. For Scotland, age-specific estimates of AAFs for certain chronic causes (e.g. cancer, hypertension, strokes and unspecified liver disease) were slightly lower by age

6 Alcohol and Alcoholism, 2016, Vol. 51, No Table 3. US death rates per 100,000 for the chronic alcohol-induced causes using only the underlying cause of death vs. using all records mentioning these causes (i.e. using multiple causes of death), for age groups and years (premature deaths), in Alcoholic liver disease (ICD-10 Code K70) Age years Age years Underlying only Any mention Underlying only Any mention Year No. Rate (CI) No. Rate (CI) No. Rate (CI) No. Rate (CI) , ( ) 15, ( ) ( ) ( ) , ( ) 18, ( ) ( ) ( ) Total 76, ( ) 100, ( ) 12, ( ) 17, ( ) APC, % (CI) APC, % (CI) APC, % (CI) APC, % (CI) 2.5 ( )* 2.6 ( )* 0.6 ( 1.2, +2.5) 1.4 ( 0.1, +2.9) Mental-behavioural disorders due to alcohol (ICD-10 F ) Year No. Rate (CI) No. Rate (CI) No. Rate (CI) No. Rate (CI) ( ) 15, ( ) ( ) ( ) ( ) 18, ( ) ( ) ( ) Total 29, ( ) 102, ( ) ( ) 20, ( ) APC, % (CI) APC, % (CI) APC, % (CI) APC, % (CI) 2.7 (1.5, 4.0)* a 2.5 (1.4, 3.7)* 2.1 ( 0.1, +4.3) 1.5 (1.5, 1.6)* a Any alcohol-induced chronic cause, b total for No. Rate (CI) No. Rate (CI) No. Rate (CI) No. Rate (CI) Total 109, ( ) 200, ( ) 19, ( ) 37, ( ) APC (CI) APC (CI) APC (CI) APC 2.4 ( )* a 2.3 ( )* 0.7 ( 0.6, +1.9) 1.1 ( )* APC, annual percent change in rate, from joinpoint regression (see text); CI, confidence interval (95%), lower and upper limits; ICD-10, International Classification of Diseases Version 10 (see text). Note: Using multiple causes of death, records with both alcoholic liver disease and mental-behavioural disorders due to alcohol would be counted in both of these categories. a Using joinpoint regression, the APCs were statistically significantly positive from 2007 to 2010 and from 2010 to 2012, with one joinpoint; the average APC for past 5 years is shown (see text). b Any alcohol-induced chronic cause, with alcohol attributable fraction (AAF) of 100%; excludes ICD-10 code F10.0 (acute alcohol intoxication) (see Table 2 and text). *P <.05 for APC or years (Information Services Division, 2009). Thus, the potential for obtaining estimated age-specific AAFs, and their impact on estimated alcohol-related deaths, should be evaluated in the US. The ARDI estimate for excessive alcohol use is 83,823 deaths from any alcohol-related underlying cause at age 20+ years in (Table 1). This is larger than the estimate of 65,000 deaths at age 18+ years in 2005 due to alcohol use disorders (Rehm et al., 2014b), which is a more restricted category, although associated with high mortality risk (Roerecke and Rehm, 2013). For any alcohol use (vs. excessive use), reported estimates have been higher, whether based on ARDI (CDC, 2013) orusingsimilarmethodsforallcauses(shield et al., 2013) or for cancers (Nelson et al., 2013). Agreement between ARDI estimates and numbers obtained in this study The method used in this study produced numbers of deaths from alcohol-related acute (external) causes as the underlying cause that were much lower than ARDI estimates (Table 1). This underestimation is relevant to premature mortality, because acute causes predominate among all alcohol-related deaths before age years (CDC, 2013), but was not unexpected. Only a small proportion of all deaths from alcohol-related external causes are by definition due to alcohol (AAF = 100%) (e.g. 4.3% at age years in ARDI) (CDC, 2013) (Table 1). Also, using multiple causes of death on death records identified <20% of US motor vehicle traffic deaths ( ) involving blood alcohol level of 0.08%+ g/dl in a national record-linkage system (Castle et al., (2014). In addition, the risk of these traffic death is also elevated for moderately elevated blood alcohol levels (<0.08 g/dl) (Heng et al., 2006). Aside from alcohol-related traffic deaths, alcohol testing of other external causes (injuries, homicides and suicides) is not routine in most US states (Kaplan et al., 2013; Hingson and Rehm, 2014). These considerations would seem to preclude adequate surveillance of US national death rates from all alcohol-related external causes. For chronic alcohol-related causes, however, using multiple-cause death records with mention of any alcohol-induced cause yielded numbers that were 81% of ARDI estimates at age years. The underestimation largely involved unspecified liver cirrhosis, cancers and strokes (each with an AAF < 100%). Combining all strokes (Table 1) is justified because heavy alcohol use is associated with increased risk of morbidity and mortality from any type of stroke (Patra et al., 2010; Zhang et al., 2014). For age 65+ years, agreement between the numbers obtained by the method used in this study vs. ARDI estimates was close only for chronic

7 60 Alcohol and Alcoholism, 2016, Vol. 51, No. 1 alcohol-related mental-nervous system causes, because almost all have an AAF of 100% (Table 1). Death certification was generally less precise (vs. younger ages), as evidenced by the larger proportion of digestive system causes that were recorded as unspecified liver cirrhosis in ARDI (Table 1). Also, deaths from mental-behavioural disorders (AAF = 100%) comprised a smaller proportion, whereas causes with AAF < 100% (e.g. unspecified liver cirrhosis, strokes and cancers) comprised a higher proportion, of all chronic alcohol-related causes at age 65+ vs years. For deaths from certain underlying causes with AAFs of <100% (e.g. AAF = 40% for unspecified liver cirrhosis, and about 1 10% for hypertension, stroke, and various cancers) (CDC, 2013), the method used in this study was of limited value (Table 1). This reflects the fact that excessive alcohol use does not always result in the diagnosis of an alcohol-induced condition. Also, certifiers may be unaware of (or choose not to mention) a decedent s history of an alcohol-induced condition. The study method, however, is an objective measure of probable current and/or past excessive alcohol use. The method yielded numbers closest to ARDI estimates (as the gold standard) for chronic alcohol-related causes at age years and for mental-nervous system causes at any age. Advantages of using multiple-cause death files in estimating alcohol-related deaths The study method using multiple-cause death records has the advantage of identifying deaths from alcohol-related infectious causes such as hepatitis C (Chen et al., 2007)notincludedinARDI.Alsoidentified are other underlying causes not defined as alcohol-related in ARDI (e.g. diabetes). Heavy drinking may increase the risk of cardiovascular events and mortality among persons with type 2 diabetes (Blomster et al., 2014; Shield et al., 2014), and also may exacerbate any chronic disease. Themethodusedinthisstudyresultedinalargernumbersof deaths vs. ARDI estimates only for hypertension at age years (but not 65+ years) (Table 1) and for IHD (not tabulated). One potential explanation for these findings is that ARDI estimates of AAFs are based on current levels of drinking in the population, whereas conditions caused by former heavy drinking may be reported on the death certificate. Another explanation, however, involves the reporting on death certificates of conditions present at death but not actually contributing to death. Certifying physicians are instructed to mention only conditions believed to have adversely affected the decedent, andif alcohol... was believed to have contributed to death then this condition should be reported even if completely unrelated to other causes (CDC, 2003). This requires a judgment by the certifier, however, and the present study did not assess actual certification practices. Future studies could include review of medical records for samples of decedents. In addition, proxy respondents can be used to assess the decedent s recent alcohol use (Tuusov et al., 2014) and drinking history. Current alcohol use could adversely affect compliance with prescribed medications for chronic diseases, thus contributing to death from chronic causes (Shield et al., 2014). Review of medical records also improves inaccuracies involved in any use of death records (including ARDI), such as under-estimating deaths from alcoholic liver disease (Manos et al., 2008) and overestimating IHD as the underlying cause (Lloyd-Jones et al., 1998; Ong et al., 2015). Surveillance of trends in US death rates from alcohol-related chronic causes Using only the underlying cause and selecting any alcohol-induced cause identified a recent increase (i.e. since 2009) in the ASMR at age years that was similar to that based on multiple causes (Table 2). This reflected the impact of the increase in the rate for alcoholic liver disease (AAF = 100%) which was the most common alcohol-induced chronic cause of death. Using the method in this study based on multiple-cause records, the recent increase in the ASMR for age years was also evident for any alcohol-related underlying cause (AAF > 0%) (Table 2), although the rates were underestimated (in view of the findings in Table 1). A recent increase was also evident using the much higher rates obtained by selecting any chronic (i.e. non-external) underlying cause including causes not defined as alcohol-related in ARDI (Table 2). These increases involved mainly alcoholic liver disease (which includes alcoholic liver cirrhosis) and liver cancer (for which cirrhosis is a precursor) as the underlying cause. Only ages <65 years have been used in some analyses of alcoholrelated deaths, because of concerns about the reliability of the underlying cause of death at older ages (Shield et al, 2013). The presence of multiple comorbid chronic diseases complicates death certification (CDC, 2003). In the present study, however, increases in the death rate at age years were evident for mental-behavioural disorders (F10 codes), and also for any alcohol-induced cause using the higher rates based on multiple causes vs. underlying cause alone (Table 3). ICD-10 coding rules do not accept F10 codes as the underlying cause if certain other alcohol-related chronic conditions (e.g. liver diseases) are also mentioned on the certificate (WHO, 2011). In conclusion, the present findings provide some support for surveillance of alcohol-induced causes based on multiple causes as well as the underlying cause alone, as also suggested for Northern Ireland (Durkin et al., 2010). Some implications for future research and for health planning Estimated AAFs in a specific population (as produced by ARDI) indicate the potential reduction in the proportion of all alcohol-related deaths that might result from interventions that change drinking patterns in the population (Eliasen et al., 2014). This includes premature deaths from chronic diseases (Nichols et al., 2012). For evaluating the actual impact of interventions, using only the underlying cause may detect temporal changes in rates for premature deaths from chronic alcohol-induced causes (Tables 2 and 3). For alcohol-related acute (external) causes of death, resources in addition to death records would be required. The larger numbers and rates based on multiple causes (vs. underlying cause alone) for chronic alcohol-related causes, however, could be included in routine mortality surveillance reports. This could be useful in assessing the burden of these deaths, for health planning and policy. Also, these data could help in addressing the public s tendency to underestimate the burden of alcohol-related deaths, while overestimating the beneficial effects of drinking on certain chronic conditions (especially for IHD) (Rehm et al., 2014a). Future studies should examine the recent increases in US rates of premature deaths related to chronic alcohol-related causes including alcoholic liver disease and its sequelae, using time-series and regression techniques (Landberg and Hubner, 2014). Analyses should include data on recent increases in US apparent per capita alcohol consumption at age 14+ years (LaVallee et al., 2014) and in the prevalence among adults of heavy and binge drinking (National Center for Health Statistics, 2014) and alcohol volume (Kerr et al., 2014). Data on increasing obesity prevalence (Kim et al., 2014) could also be included. Only short lag times may be needed, however, for changes in drinking habits to affect mortality from cirrhosis (Shield et al., 2014). The cirrhosis death rate based on the underlying cause is the

8 Alcohol and Alcoholism, 2016, Vol. 51, No indicator of alcohol-related mortality used for surveillance and health policy in the US. (Crombie et al., 2007). CONFLICTS OF INTEREST STATEMENT None declared. REFERENCES Blomster JI, Zoungas S, Chalmers J, et al. (2014) The relationship between alcohol consumption and vascular complications and mortality in individuals with type 2 diabetes. Diabetes Care 37: Castle IP, Yi H, Hingson RW, et al. (2014) State variation in underreporting of alcohol involvement on death certificates: Motor vehicle traffic crash fatalities as an example. J Stud Alcohol Drugs 75: Centers for Disease Control and Prevention. (2003) Physicians Handbook on Medical Certification of Death Revision. Hyattsville, MD: Department of Health and Human Services (DHHS) Publication No. (PHS) Centers for Disease Control and Prevention. (2013) Alcohol-related disease impact (ARDI) application. aspx. Centers for Disease Control and Prevention. (2014a). National Center for Health Statistics. Multiple cause of death file on CDC WON- DER online database, released html. (17 December 2014, date last accessed). Centers for Disease Control and Prevention. (2014b) Alcohol-attributable deaths and years of potential life lost 11 states, MMWR Morb Mortal Wkly Rep 63: Chen CM, Yoon YH, Yi HY, et al. (2007) Alcohol and hepatitis C mortality among males and females in the United States: A life-table analysis. Alcohol Clin Exp Res 31: Clegg LX, Hankey BF, Tiwari R, et al. (2009) Estimating average annual per cent change in trend analysis. Stat Med 28: Crombie IK, Irvine L, Elliott L, et al. (2007) How do public health policies tackle alcohol-related harm: A review of 12 developed countries. Alcohol Alcohol 42: Durkin A, Connolly S, O Reilly D. (2010) Quantifying alcohol-related mortality: Should alcohol-related contributory causes of death be included? Alcohol Alcohol 45: Eliasen M, Becker U, Grennbaek M, et al. (2014) Alcohol-attributable and alcohol-preventable mortality in Denmark: An analysis of which intake levels contribute most to alcohol s harmful and beneficial effects. Eur J Epidemiol 29: Gonzales K, Roeber J, Kanny D, et al. (2014) Alcohol-attributable deaths and years of potential life lost 11 states, MMWR Morb Mortal Wkly Rep 63: Heng K, Hargarten S, Layde P, et al. (2006) Moderate alcohol intake and motor vehicle crashes: The conflict between health advantage and at-risk use. Alcohol Alcohol 41: Hingson R, Rehm J. (2014) Measuring the burden: Alcohol s evolving impact. Alcohol Res 35: Information Services Division. (2009) Alcohol Attributable Mortality and Morbidity June Edinburgh, Scotland: NHS National Services Scotland. alcoholfractions.rep.pdf. (14 July 2014, date last accessed). Johnson NA, Kypri K, Latter J, et al. (2014) Prevalence of unhealthy alcohol use on hospital outpatients. Drug Alcohol Depend 144: Kaplan MS, McFarland BH, Huguet N, et al. (2013) Acute alcohol intoxication and suicide: a gender-stratified analysis of the National Violent Death Reporting System. Inj Prev 19: Kerr WC, Mulia N, Zemore SE. (2014) U.S. trends in light, moderate, and heavy drinking episodes from 2000 to Alcohol Clin Exp Res 38: Kim HJ, Fay MP, Feuer EJ, et al. (2000) Permutation tests for joinpoint regression with applications to cancer rates. Statist Med 19: Kim Y, Ejaz A, Tayal A, et al. (2014) Temporal trends in population-based death rates associated with chronic liver disease and liver cancer in the United States over the last 30 years. Cancer 120: Knott CS, Coombs N, Stamatakis E, et al. (2015) All cause mortality and the case for age specific alcohol consumption guidelines: Pooled analyses of up to 10 population based cohorts. BMJ 350:h384. Landberg J, Hubner L. (2014) Changes in the relationship between volume of consumption and alcohol-related problems in Sweden during Alcohol Alcohol 49: LaVallee RA, Kim T, Yi H. (2014) Apparent per capita alcohol consumption: National, state, and regional trends, Surveillance report #98. National Institute on Alcohol Abuse and Alcoholism. U.S. Department of Health and Human Services. Lloyd-Jones DM, Martin DO, Larson MG, et al. (1998) Accuracy of death certificates for coding coronary heart disease as the cause of death. Ann Intern Med 129: Manos MM, Leyden WA, Murphy RC, et al. (2008) Limitations of conventionally derived chronic liver disease mortality rates: Results of a comprehensive assessment. Hepatology 47: Martin J, Barry J, Goggin D, et al. (2010) Alcohol-attributable mortality in Ireland. Alcohol Alcohol 45: Moyer LA, Boyle CA, Pollock DA. (1989) Validity of death certificates for injury-related causes of death. Am J Epidemiol 130: National Center for Health Statistics (2014) Health United States, 2013: With special feature on prescription drugs. Hyattsville, MD. Nelson DE, Jarman DW, Rehm J, et al. (2013) Alcohol-attributable cancer deaths and years of potential life lost in the United States. Am J Public Health 103: Nichols M, Scarboroush P, Allender S, et al. (2012) What is the optimal level of population alcohol consumption for chronic disease prevention in England? Modelling the impact of changes in average consumption levels. BMJ Open 2:e Ong P, Gambatese M, Begler E, et al. (2015) Effect of cause-of-death training on agreement between hospital discharge diagnoses and cause of death reported, inpatient hospital deaths, New York City, Prev Chron Dis 14:E04. Patra J, Taylor B, Irving H, et al. (2010) Alcohol consumption and the risk of morbidity and mortality for different stroke types a systematic review and meta-analysis. BMC Public Health 10:258. Peto R, Doll R. (1992) Smoking accepted on death certificates. BMJ 305: Pollock DA, Boyle CA, DeStefano F, et al. (1987) Underreporting of alcoholrelated mortality in death certificates of young US Army veterans. JAMA 258: Rehm J, Shield KD. (2014) Global alcohol-attributable deaths from cancer, liver cirrhosis, and injury in Alcohol Res 35: Rehm J, Lachenmeier DW, Room R. (2014a) Why does society accept a higher risk for alcohol than for other voluntary or involuntary risks? BMC Medicine 12:189. Rehm J, Dawson D, Frick U, et al. (2014b) Burden of disease associated with alcohol use disorders in the United States. Alcohol Clin Exp Res 38: Roerecke M, Rehm J. (2013) Alcohol use disorders and mortality: a systematic review and meta-analysis. Addiction 108: Roerecke M, Greenfield K, Kerr WC, et al. (2011) Heavy drinking occasions in relation to ischemic heart disease mortality an year follow-up of the 1984 and 1995 US National Alcohol Surveys. Int J Epidemiol 40: Saitz R, Freedner N, Palfai TP, et al. (2006) The severity of unhealthy alcohol use in hospitalized medical patients: the spectrum is narrow. J Gen Intern Med 21: Shield KD, Gmel G, Kehoe-Chan T, et al. (2013) Mortality and potential years of life lost attributable to alcohol consumption by race and sex in the United States, PLOS One 8:E Shield KD, Parry C, Rehm J. (2014) Chronic diseases and conditions related to alcohol use. Alcohol Res 35:

9 62 Alcohol and Alcoholism, 2016, Vol. 51, No. 1 Stahre M, Roeber J, Kanny D, et al. (2014) Contribution of excessive alcohol consumption to deaths and years of potential life lost in the United States. Prev Chronic Dis 11: Tuusov J, Lang K, Vali M, et al. (2014) Prevalence of alcohol-related pathologies at autopsy: Estonian forensic study of alcohol and premature death. Addiction 109: World Health Organization. (2011) International Classification of Diseases and Related Health Problems, Tenth Revision. Volume 2, Instruction Manual. Edition Geneva: Switzerland: WHO. Zhang C, Qin YY, Chen Q, et al. (2014) Alcohol intake and risk of stroke: a dose-response meta-analysis of prospective studies. Int J Cardiol 174: