Eur J Clin Pharmacol (2015) 71:637–642 DOI 10.1007/s00228-015-1841-z
PHARMACOEPIDEMIOLOGY AND PRESCRIPTION
Association between hypnotics use and increased mortality: causation or confounding? C. Ineke Neutel 1 & Helen L. Johansen 1
Received: 20 November 2014 / Accepted: 24 March 2015 / Published online: 7 April 2015 # Springer-Verlag Berlin Heidelberg 2015
Abstract Purpose Many research studies have found associations between benzodiazepines and/or z-hypnotics (BZZ) and increasing mortality, leading to a discussion about causation or confounding. This study suggests a factor that could produce this association through confounding. Methods The Norwegian population in 2010 supplied 8862 deaths ages 41–80 and 898,289 controls. Index dates were added to control records which corresponded to death dates. BZZ use was recorded for 2 years before death/index date. Results Persons exposed to BZZ were more likely (OR=2.3) to die than those who were not. With proximity of death, increasingly larger proportions of the prospective deaths received prescriptions for BZZ, until in the last 2 months 40– 45 % received BZZ. The frequency of BZZ use in controls increased with age as opposed to the death cohort where all ages showed similar rates of BZZ use. In the last few months before death, the youngest age group had an OR=5.8 for BZZ use while the oldest age group an OR=1.8, adjusted for age and sex. Opioid use showed a similar pattern of increasing use near death. Conclusions The increased use of BZZ with approaching death is consistent with increasing symptomatic treatment in terminal illness. Thus, the association of BZZ and mortality is more likely to be due to confounding than to causality. Further evidence from this and other research includes similar use patterns for other drugs such as opioids, the lack of specificity
* C. Ineke Neutel [email protected] 1
School of Epidemiology, Public Health and Preventive Medicine, University of Ottawa, 451 Smyth Road (Room 3105) Roger-Guindon Building, Ottawa K1H 8M5, Ontario, Canada
in cause of death and the size of the association regarding age and time to death. Keywords Hypnotics . Benzodiazepines . z-Hypnotics . Terminal illness . Mortality . Causation . Confounding
Introduction Several studies found associations between exposure to hypnotics such as benzodiazepines and/or z-hypnotics (BZZ) and increased mortality [1–8]. Even after adjusting for many potential confounders, such as such as age, sex, smoking, BMI, marital status, health status, ethnicity, alcohol, prior cancer, etc., an increase in all-cause mortality remained for those who had used hypnotics [1]. Failing to see other reasons for this association, many concluded it was causal. The association was found not only for hypnotic use and all-cause mortality [1–4] but also for more specific causes of death such as cardiovascular disease [2, 5, 6], diabetes [2], kidney disease [7], and cancer deaths [1, 9]. Other studies did not find an association between use of hypnotics and increased mortality [10–12]. In his paper BDo no harm: not even to some degree,^ Kripke stated BIt would be wonderful if somebody could prove it [the association being causal] was not so…^ [13]. Unfortunately, it is difficult to prove that an increased risk does not exist, especially when so many studies have shown an association between BZZ and mortality. The most convincing evidence would be to show the existence of an alternative explanation. This article aims to present a confounding factor for which none of these studies have been adjusted. Our hypothesis is that the association of BZZ use and mortality can be explained, partially or wholly, by increased BZZ use by the terminally ill.
638
Eur J Clin Pharmacol (2015) 71:637–642
Methodology The study population was derived from the Norwegian Pharmacy Database (NorPD) in which all prescriptions dispensed in Norwegian pharmacies are registered [14, 15]. Information on prescription records include person identification (id), prescriber id, age, sex, medication used, and date of death. Medications were coded by the Anatomical Therapeutic Chemical (ATC) classification [16]. ATC codes selected for this study include benzodiazepines (ATC: N05BA, N05CD), z-hypnotics (ATC: N05CF), and opioids (ATC: N02A). Excluded are prescriptions without person id recorded which amounts to about 0.5 % of prescriptions. The study population comprised all persons in Norway, aged 4180, who were dispensed any (or several) of the three mentioned types of drugs during the study period. Combining prescriptions for each person provided a person-based population, aged 41–80 in 2010. The age of 80 was used as a cutoff because of the increasing institutionalization with increasing age and that no information was available on drug use in institutions. Deaths were excluded except for those persons who died in 2010. The design of this study is that of a case-control study with cases who died in the first 10 months of 2010 and controls that were still alive by the end of 2010. BZZ use will be measured in two ways: (1) whether or not BZZ was dispensed at any time during the 24-month exposure period to members of case and control groups and (2) whether or not BZZ was dispensed in each 2-month sub periods over the 24-month exposure time. All logistic regression models were adjusted for age in single years and sex. The exposure time consisted of the 2 years before the 2month period during which the death/index date occurred. Since deaths occurred over a period of 10 months, the time period during which the deaths occurred, January to October 2010, was divided into five 2-month sub periods. A random number from 1 to 5 was inserted on each record of the control population to be used as an index period comparable to one of the death periods. For example, index number 1 would refer to the months January and February, index number 2 to March and April and so on (see chart). For both control group and death cohort BZZ prescriptions were recorded over a two year period before the index/ death period (see chart). Thus, comparable case and control groups were achieved. Index/death period 1. January to February 2010 2. March to April, 2010 3. May to June, 2010 4. July to August, 2010 5. September to October, 2010
24-month exposure periods January 2008 to December, 2009 March 2008 to February, 2010 May 2008 to April 2010 July 2008 to June 2010 September 2008 to August 2010
All exposure periods were combined and exposure time was expressed in terms of months prior to the relevant index/death period. For statistical analysis, SAS version 9.2 (SAS Institute, Cary, NC, USA) was used.
Results The death cohort included 9491 people, who died between January and October, 2010 (Table 1). The control group contained 898,289 people, who were still alive by the end of 2010. The death cohort included more males (56.6 %) while the control group included more females (56.7 %). The controls on the whole were younger than the deaths. Table 2a presents the association between BZZ use and mortality in the death cohort versus the control group with mortality as dependent variable. This table shows that over the 24 months of the exposure period, the persons in the death cohort were more likely to receive BZZ prescriptions (OR= 2.3) than were the controls. Opioid use over the same time period showed a similar pattern of use. Likelihood of opioid use by the death cohort is 1.6 times that of the control group when adjusted for age and sex (Table 2b). The frequency of BZZ use varied with age and time to death (Fig. 1). The proportion of controls receiving BZZ increased with age from 13 % of the 41–50 year age group up to 27 % of the 71–80-year age group. The death cohort received more BZZ than controls at all ages, with an increasing proportion of users as death nears. Already by 24 months before death, the prospective deaths received more prescriptions for BZZ than did the controls for every age group. This excess was the greatest for the younger age groups where the BZZ use in the death cohort was already double the BZZ use in the control group, while the increase was much smaller for older age groups. BZZ use for the death cohort increased gradually over the 24 months of the exposure period. In the last 2-month period prior to death, the 41–50 year olds in the death cohort were more than three times as likely to receive BZZ compared to the controls of the same age group, while for the oldest age group, the likelihood of use was almost 50 % higher. Figure 2 presents logistic regression models for each 2month period of the exposure time prior to death. Each OR is a separate model comparing deaths to controls for each 2-month time period and adjusted for sex and age in single years. When all ages are combined, the ORs for both BZZ and opioids are the highest for the time period nearest date of death at (ORs 2.5 and 2.2) and lowest at the beginning of the exposure period, 2 years earlier (OR 1.4) (Fig. 2a). Age-specific series of models were produced for 10-year age groups, also adjusted for age in single years within the larger age groups (Fig. 2b). The youngest age group showed much greater increases in BZZ use (OR 5.8) prior
Eur J Clin Pharmacol (2015) 71:637–642 Table 1 Attributes of control group and death cohort at death or index date
639
Variables
Controls
Total Sex Age
No. of BZZ prescriptions over the 2-year exposure time
No. of opioid prescriptions over the 2-year exposure time
male female 41–50 51–60 61–70 71–80 0 1–5 5–20 21–40 41+ 0 1–5 5–20 21–40 41+
Prospective deaths
N
%
N
%
898,289
100
9491
100
388,884 509,405 250,271 258,669 233,325 156,024 453,360 270,184 130,556 29,362 14,827 450,944 368,957 55,442 16,599 6347
43.3 56.7 27.9 28.8 26.0 17.4 50.5 30.1 14.5 3.3 1.7 50.2 41.1 6.2 1.8 0.7
5375 4116 581 1434 2980 4496 2617 3172 2540 710 452 3708 4113 1202 338 130
56.6 43.4 6.1 15.1 31.4 47.4 27.6 33.4 26.8 7.5 4.8 39.1 43.3 12.7 3.6 1.4
Controls are from the Norwegians pharmaceutical data base (NorPD) and between ages 41-80 in 2010 and alive by the end of 2010 Prospective deaths include all those between ages 41-80 who died between January 2010 and October 2010
to death than did the older age groups in the last few months (OR 1.8) for the oldest.
Discussion A considerable number of studies, including the present one, found an association between BZZ use and increased mortality [1–8]. The question is whether this association can be considered causal or whether an unadjusted confounder is Table 2 Odds ratio (OR) for BZZ use and for opioid use, both adjusted for age and sex Variable
Reference
a. BZZ use and mortality Sex Females Age in 10 years Age 41–50 BZZ usea No BZZ b. Opioid use and mortality Sex Females Age in 10 years Age 41–50 Opioid useb No opioids
OR
Confidence interval
0.47 2.20 2.30
0.45 2.16 2.20
0.49 2.25 2.40
0.54 2.36 1.62
0.52 2.31 1.56
0.56 2.41 1.69
a
BZZ use at any time over the 2-year exposure period prior to death/index period
b Opioid use at any time over the 2-year exposure period prior to the index/death date
present. The present study illustrated that the frequency of filling BZZ prescriptions increased with approaching death, culminating with the greatest frequency in the last few months before death. This pattern of gradually increasing BZZ accessibility until time of death is consistent with a gradual increase in symptomatic treatment of the increasing pain and other discomforts of approaching death. This pattern of BZZ availability would show itself as an association between BZZ use and increasing mortality, and may well be the confounder underlying the association found in this and many other studies. Opioid use which has not been implicated in mortality in the same way as BZZ has been was shown to have a similar association with mortality. Thus, we conclude that the association between BZZ availability and mortality can be at least be partially, and possibly be wholly, explained by an increase in symptomatic treatment of increasing discomfort as death draws near. As with all studies, this study has strengths and limitations. Strengths include the availability of the data for the entire Norwegian population, as well as confidence in the reliability of the data. Strengths also include the information available on each prescription, such as prescriber id, age, sex, ATC code, and date of death. One concern is the lack of information on the institutionalized since their medications could not be recorded in the NorPD. This is also the reason for the age cutoff at age 80 since institutionalization increases with age. In addition, many people are hospitalized at least some of the
640
Eur J Clin Pharmacol (2015) 71:637–642
Fig. 1 Percentage of population using BZZ prior to death/index date and age. Controls were derived from the Norwegians pharmaceutical data base (NorPD) and were between ages 41–80 in 2010 and alive by the end of 2010. Prospective deaths include all those in the NorPD who were between ages 41 and 80 in 2010 and who died between January 2010 and October 2010
time shortly before death, and one can assume that the persons in the death cohort were more likely to be hospitalized than the control group. Thus, they likely received more BZZ than is recorded. As a result, the increased BZZ use in the death cohort as indicated in this study may be considered an underestimate. Another potential ambiguity is the day of death within the 2-month death period. The 2-year exposure period was calculated backward from the 2-month period during which the date of death or the index date occurred. Thus, for deaths, there is a period of between a minimum of 1 day and a maximum of 60 days prior to the day of death for which no exposure was measured for this study. The advantage of this lag period is that some of the irregularities of the last few weeks of terminal illness, such as the increased potential for institutionalization or palliative sedation, are less likely to affect the data. A disadvantage of this lag period is that the number of days between beginning of the period and death is variable. For roughly one quarter of deaths, the lag period will be less than 2 weeks, but for about three quarters of deaths, the lag period will be more than 2 weeks. On average, we can expect this lag time to reduce the irregularities in BZZ use of the last while before death and, on average, we can expect this lag in Fig. 2 Odds ratios (OR) for BZZ use by months before index/death period adjusted for age and sex. a For all ages and both BZZ and opioids. b For BZZ stratified by age groups. See legend of Fig. 1 for description of controls and prospective deaths. Logistic regression models prospective deaths and controls were calculated for each point separately and adjusted for sex and age in single years even within the larger age group
exposure time to reduce the association between BZZ use and mortality, making the results of this study a more conservative estimate. Based on the pattern of BZZ use in the months before death, we concluded that the pattern of BZZ prescribing in the death cohort is consistent with treatment of increasing symptoms of pain and discomfort in the final months of life. It should not be a surprise that more of these types of medications are used in the last year of life [17, 18]. This pattern of use is most likely enough to explain the overall association between BZZ use and mortality found in this study and also that found in other studies. Others who found an association between BZZ and increased mortality have come to different conclusions. Kripke et al. concluded that the association was most likely causal, especially for cancer [1, 9, 13]. Major research projects were carried out, adjusting for many variables such as age, sex, smoking, BMI, marital status, health status, ethnicity, alcohol, and prior cancer, in order to adjust for potential confounding. Based on the positive association of BZZ and mortality remaining after adjusting for many potential confounding, Kripke et al. concluded that BZZ are dangerous drugs [9]. Although numerous variables were measured and adjusted
a) For all ages and both BZZ and opioids.
b) For BZZ stratified by age groups
Eur J Clin Pharmacol (2015) 71:637–642
for, what could neither be measured nor adjusted for were the increasingly severe symptoms of pain and other discomfort that often herald impending death, since information on severity of symptoms, other than that relating to number of conditions from which a person suffers, is rarely available. This variable then has become a confounder. A variety of other evidence also supports confounding rather than causality. One factor is the lack of specificity in the cause of death showing the association with BZZ, ranging from all-cause mortality to cancer to cardiovascular disease, to kidney disease, etc. [1–7, 9]. It is difficult to think of a biological reason for increased risk of mortality that is true for such a wide range of conditions. Other evidence for confounding also comes from similar patterns of increased BZZ with age and time to death. The present study shows that the youngest age group (41–50) shows a much larger association between BZZ and mortality than the oldest age group (71–80). Among published studies, those with an older study populations, e.g. over 65, show little association between BBZ and mortality [6, 12, 19, 20], while those with study populations with people under 60 are more likely to show an association [1, 2, 21]. Another factor shown to determine the size of the association between BZZ and mortality in the studies cited is the amount of time between the recording of BZZ use and death. The present study shows already a 50 % increase in use of BZZ by 24 months before death. Many serious diseases leading to terminal illness have uncomfortable symptoms such as pain or difficulty sleeping already several years before death [22, 23]; thus, the gradual increase in BZZ use starting years before death is not surprising. The association of BZZ and mortality was shown to increase with proximity of death. Thus, such an association is more likely to occur and be higher, if at least some of the study population has only a short time left till death. Of the studies showing an association between BZZ use and mortality, Kripke et al.’s 2012 study shows one of the highest OR but it also had one of the shortest time lapses between the measurement of BZZ exposure and detection of death, i.e. an average of 2.5 years [1]. This short exposure time as well as the relatively young study population would be expected to lead to a large association if the increases of BZZ use are due to the increase in of symptoms as death approaches. Other evidence comes from the use of other medications under similar circumstances; particularly, other medications use to relieve symptoms of severe illnesses are likely to occur prior to death. While opioids are known to have the potential to hasten death in cases of severe pain and high tolerance, that is different from the way BZZ have been considered dangerous even when used when the person is still healthy [9]. Still, this study shows opioids to have a similar association with mortality as BZZ use. Other medications with greater use before death were antidepressants [24] and antipsychotics
641
[25] and an increase in various drugs of symptom management [18]. Surely, we cannot consider that all of these drugs are dangerous and to be avoided because they appear to have an association with increasing the risk of death. Confounding by indication or reason for prescription [26] or reverse causation [27, 28] seems to be more plausible. In conclusion, the pattern of increasing BZZ use before death is consistent with its use as treatment of worsening symptoms with approaching death. Much evidence has been presented here supporting the notion that the association between BZZ and mortality is at least partially due to confounding, rather than being instrumental in initiating new disease leading to mortality. Whether the association between BZZ use and mortality is considered causal or confounding is an important distinction since the interpretation may determine how physicians use the medication. While some of Kripke’s message about the use of BZZ needs to be taken to heart [1, 9, 13, 29] and physicians need to be cautioned regarding the overuse or inappropriate use of BZZ, it is also true that when people are very uncomfortable when suffering from anxiety or lack of sleep, physicians should not avoid using BZZ for fear it increases the risk of death. Acknowledgments The authors would like to thank Drs. Svetlana Skurtveit, Christian Berg, and colleagues of the Norwegian Institute of Public Health, Oslo, Norway, for their support. Conflict of interest Neither author has any conflict of interest associated with any part of this study and article. Author involvement Both authors were involved at all stages of the research and writing of the article. While Dr. Neutel took the lead, Dr. Johansen was also involved in planning of the research methodology, the analysis of the data, and the editing of the paper.
References 1.
2. 3.
4.
5.
6.
Kripke DF, Langer RD, Kline LE (2012) Hypnotics' association with mortality or cancer: a matched cohort study. BMJ Open 2(1): e000850 Mallon L, Broman JE, Hetta J (2009) Is usage of hypnotics associated with mortality? Sleep Med 10(3):279–286 Belleville G (2010) Mortality hazard associated with anxiolytic and hypnotic drug use in the National Population Health Survey. Can J Psychiatry 55(9):558–567 Hausken AM, Skurtveit S, Tverdal A (2007) Use of anxiolytic or hypnotic drugs and total mortality in a general middle-aged population. Pharmacoepidemiol Drug Saf 16(8):913–918 Mallon L, Broman JE, Hetta J (2002) Sleep complaints predict coronary artery disease mortality in males: a 12-year follow-up study of a middle-aged Swedish population. J Intern Med 251(3): 207–216 Merlo J et al (1996) Increased risk of ischaemic heart disease mortality in elderly men using anxiolytics-hypnotics and analgesics. Results of the 10-year follow-up of the prospective population study "Men born in 1914", Malmo, Sweden. Eur J Clin Pharmacol 49(4):261–265
642 7.
8.
9. 10. 11.
12. 13.
14.
15.
16.
17.
18.
Eur J Clin Pharmacol (2015) 71:637–642 Fukuhara S et al (2006) Symptoms of depression, prescription of benzodiazepines, and the risk of death in hemodialysis patients in Japan. Kidney Int 70(10):1866–1872 Weich S et al (2014) Effect of anxiolytic and hypnotic drug prescriptions on mortality hazards: retrospective cohort study. BMJ 348:g1996 Kripke DF (2009) Do hypnotics cause death and cancer? The burden of proof. Sleep Med 10(3):275–276 Phillips B, Mannino DM (2005) Does insomnia kill? Sleep 28(8): 965–971 Ahmad R, Bath PA (2005) Identification of risk factors for 15-year mortality among community-dwelling older people using Cox regression and a genetic algorithm. J Gerontol A Biol Sci Med Sci 60(8):1052–1058 Jaussent I et al (2013) Hypnotics and mortality in an elderly general population: a 12-year prospective study. BMC Med 11:212 Kripke DF, Langer RD, Kline LE (2012) Do no harm: not even to some degree. J Clin Sleep Med JCSM Off Publ Am Acad Sleep Med 8(4):353–354 Furu K (2001) Drug utilisation in a public health perspective: Establishing a national prescription register in Norway. Nor J Epidemiol 11:55–60 Furu K et al (2010) The Nordic countries as a cohort for pharmacoepidemiological research. Basic Clin Pharmacol Toxicol 106(2):86–94 WHO, Collaborating Centre for Drugs Statistics Methodology: ATC/DDD Index 2011: Oslo, Norway, available from: http:// www.whocc.no last accessed 10/05/2013 Fahlman C et al (2006) Prescription drug spending for Medicare + Choice beneficiaries in the last year of life. J Palliat Med 9(4):884– 893 Langton JM (2013) Prescription drug use in the last 12 months of life: observations in a cohort of elderly patients, in International
Conference of Pharmacoepidemiology. Pharmacoepidemiol Drug Saf Montr. Abstract 703 19. Chen HC, Su TP, Chou P (2013) A nine-year follow-up study of sleep patterns and mortality in community-dwelling older adults in Taiwan. Sleep 36(8):1187–1198 20. Hays JC, Blazer DG, Foley DJ (1996) Risk of napping: excessive daytime sleepiness and mortality in an older community population. J Am Geriatr Soc 44(6):693–698 21. Kojima M et al (2000) Sleep patterns and total mortality: a 12-year follow-up study in Japan. J Epidemiol 10(2):87–93 22. Coleman MP et al (2008) Cancer survival in five continents: a worldwide population-based study (CONCORD). Lancet Oncol 9(8):730–756 23. Verdecchia A et al (2008) Patient survival for all cancers combined as indicator of cancer control in Europe. Eur J Public Health 18(5): 527–532 24. Brelin S et al (2013) Antidepressants to cancer patients during the last year of life— population-based study. Psychooncology 22(3): 506–514 25. Wang PS et al (2005) Risk of death in elderly users of conventional vs. atypical antipsychotic medications. N Engl J Med 353(22): 2335–2341 26. Strom B (2005) Pharmacoepidemiology. John Wiley & Sons Ltd., Chichester 27. Pottegard A et al (2013) Use of benzodiazepines or benzodiazepine related drugs and the risk of cancer: a population-based case-control study. Br J Clin Pharmacol 75(5):1356–1364 28. Kao CH et al (2012) Relationship of zolpidem and cancer risk: a Taiwanese population-based cohort study. Mayo Clin Proc 87(5): 430–436 29. Kripke DF (2007) Greater incidence of depression with hypnotic use than with placebo. BMC Psychiatry 7:42
PHARMACOEPIDEMIOLOGY AND PRESCRIPTION
Association between hypnotics use and increased mortality: causation or confounding? C. Ineke Neutel 1 & Helen L. Johansen 1
Received: 20 November 2014 / Accepted: 24 March 2015 / Published online: 7 April 2015 # Springer-Verlag Berlin Heidelberg 2015
Abstract Purpose Many research studies have found associations between benzodiazepines and/or z-hypnotics (BZZ) and increasing mortality, leading to a discussion about causation or confounding. This study suggests a factor that could produce this association through confounding. Methods The Norwegian population in 2010 supplied 8862 deaths ages 41–80 and 898,289 controls. Index dates were added to control records which corresponded to death dates. BZZ use was recorded for 2 years before death/index date. Results Persons exposed to BZZ were more likely (OR=2.3) to die than those who were not. With proximity of death, increasingly larger proportions of the prospective deaths received prescriptions for BZZ, until in the last 2 months 40– 45 % received BZZ. The frequency of BZZ use in controls increased with age as opposed to the death cohort where all ages showed similar rates of BZZ use. In the last few months before death, the youngest age group had an OR=5.8 for BZZ use while the oldest age group an OR=1.8, adjusted for age and sex. Opioid use showed a similar pattern of increasing use near death. Conclusions The increased use of BZZ with approaching death is consistent with increasing symptomatic treatment in terminal illness. Thus, the association of BZZ and mortality is more likely to be due to confounding than to causality. Further evidence from this and other research includes similar use patterns for other drugs such as opioids, the lack of specificity
* C. Ineke Neutel [email protected] 1
School of Epidemiology, Public Health and Preventive Medicine, University of Ottawa, 451 Smyth Road (Room 3105) Roger-Guindon Building, Ottawa K1H 8M5, Ontario, Canada
in cause of death and the size of the association regarding age and time to death. Keywords Hypnotics . Benzodiazepines . z-Hypnotics . Terminal illness . Mortality . Causation . Confounding
Introduction Several studies found associations between exposure to hypnotics such as benzodiazepines and/or z-hypnotics (BZZ) and increased mortality [1–8]. Even after adjusting for many potential confounders, such as such as age, sex, smoking, BMI, marital status, health status, ethnicity, alcohol, prior cancer, etc., an increase in all-cause mortality remained for those who had used hypnotics [1]. Failing to see other reasons for this association, many concluded it was causal. The association was found not only for hypnotic use and all-cause mortality [1–4] but also for more specific causes of death such as cardiovascular disease [2, 5, 6], diabetes [2], kidney disease [7], and cancer deaths [1, 9]. Other studies did not find an association between use of hypnotics and increased mortality [10–12]. In his paper BDo no harm: not even to some degree,^ Kripke stated BIt would be wonderful if somebody could prove it [the association being causal] was not so…^ [13]. Unfortunately, it is difficult to prove that an increased risk does not exist, especially when so many studies have shown an association between BZZ and mortality. The most convincing evidence would be to show the existence of an alternative explanation. This article aims to present a confounding factor for which none of these studies have been adjusted. Our hypothesis is that the association of BZZ use and mortality can be explained, partially or wholly, by increased BZZ use by the terminally ill.
638
Eur J Clin Pharmacol (2015) 71:637–642
Methodology The study population was derived from the Norwegian Pharmacy Database (NorPD) in which all prescriptions dispensed in Norwegian pharmacies are registered [14, 15]. Information on prescription records include person identification (id), prescriber id, age, sex, medication used, and date of death. Medications were coded by the Anatomical Therapeutic Chemical (ATC) classification [16]. ATC codes selected for this study include benzodiazepines (ATC: N05BA, N05CD), z-hypnotics (ATC: N05CF), and opioids (ATC: N02A). Excluded are prescriptions without person id recorded which amounts to about 0.5 % of prescriptions. The study population comprised all persons in Norway, aged 4180, who were dispensed any (or several) of the three mentioned types of drugs during the study period. Combining prescriptions for each person provided a person-based population, aged 41–80 in 2010. The age of 80 was used as a cutoff because of the increasing institutionalization with increasing age and that no information was available on drug use in institutions. Deaths were excluded except for those persons who died in 2010. The design of this study is that of a case-control study with cases who died in the first 10 months of 2010 and controls that were still alive by the end of 2010. BZZ use will be measured in two ways: (1) whether or not BZZ was dispensed at any time during the 24-month exposure period to members of case and control groups and (2) whether or not BZZ was dispensed in each 2-month sub periods over the 24-month exposure time. All logistic regression models were adjusted for age in single years and sex. The exposure time consisted of the 2 years before the 2month period during which the death/index date occurred. Since deaths occurred over a period of 10 months, the time period during which the deaths occurred, January to October 2010, was divided into five 2-month sub periods. A random number from 1 to 5 was inserted on each record of the control population to be used as an index period comparable to one of the death periods. For example, index number 1 would refer to the months January and February, index number 2 to March and April and so on (see chart). For both control group and death cohort BZZ prescriptions were recorded over a two year period before the index/ death period (see chart). Thus, comparable case and control groups were achieved. Index/death period 1. January to February 2010 2. March to April, 2010 3. May to June, 2010 4. July to August, 2010 5. September to October, 2010
24-month exposure periods January 2008 to December, 2009 March 2008 to February, 2010 May 2008 to April 2010 July 2008 to June 2010 September 2008 to August 2010
All exposure periods were combined and exposure time was expressed in terms of months prior to the relevant index/death period. For statistical analysis, SAS version 9.2 (SAS Institute, Cary, NC, USA) was used.
Results The death cohort included 9491 people, who died between January and October, 2010 (Table 1). The control group contained 898,289 people, who were still alive by the end of 2010. The death cohort included more males (56.6 %) while the control group included more females (56.7 %). The controls on the whole were younger than the deaths. Table 2a presents the association between BZZ use and mortality in the death cohort versus the control group with mortality as dependent variable. This table shows that over the 24 months of the exposure period, the persons in the death cohort were more likely to receive BZZ prescriptions (OR= 2.3) than were the controls. Opioid use over the same time period showed a similar pattern of use. Likelihood of opioid use by the death cohort is 1.6 times that of the control group when adjusted for age and sex (Table 2b). The frequency of BZZ use varied with age and time to death (Fig. 1). The proportion of controls receiving BZZ increased with age from 13 % of the 41–50 year age group up to 27 % of the 71–80-year age group. The death cohort received more BZZ than controls at all ages, with an increasing proportion of users as death nears. Already by 24 months before death, the prospective deaths received more prescriptions for BZZ than did the controls for every age group. This excess was the greatest for the younger age groups where the BZZ use in the death cohort was already double the BZZ use in the control group, while the increase was much smaller for older age groups. BZZ use for the death cohort increased gradually over the 24 months of the exposure period. In the last 2-month period prior to death, the 41–50 year olds in the death cohort were more than three times as likely to receive BZZ compared to the controls of the same age group, while for the oldest age group, the likelihood of use was almost 50 % higher. Figure 2 presents logistic regression models for each 2month period of the exposure time prior to death. Each OR is a separate model comparing deaths to controls for each 2-month time period and adjusted for sex and age in single years. When all ages are combined, the ORs for both BZZ and opioids are the highest for the time period nearest date of death at (ORs 2.5 and 2.2) and lowest at the beginning of the exposure period, 2 years earlier (OR 1.4) (Fig. 2a). Age-specific series of models were produced for 10-year age groups, also adjusted for age in single years within the larger age groups (Fig. 2b). The youngest age group showed much greater increases in BZZ use (OR 5.8) prior
Eur J Clin Pharmacol (2015) 71:637–642 Table 1 Attributes of control group and death cohort at death or index date
639
Variables
Controls
Total Sex Age
No. of BZZ prescriptions over the 2-year exposure time
No. of opioid prescriptions over the 2-year exposure time
male female 41–50 51–60 61–70 71–80 0 1–5 5–20 21–40 41+ 0 1–5 5–20 21–40 41+
Prospective deaths
N
%
N
%
898,289
100
9491
100
388,884 509,405 250,271 258,669 233,325 156,024 453,360 270,184 130,556 29,362 14,827 450,944 368,957 55,442 16,599 6347
43.3 56.7 27.9 28.8 26.0 17.4 50.5 30.1 14.5 3.3 1.7 50.2 41.1 6.2 1.8 0.7
5375 4116 581 1434 2980 4496 2617 3172 2540 710 452 3708 4113 1202 338 130
56.6 43.4 6.1 15.1 31.4 47.4 27.6 33.4 26.8 7.5 4.8 39.1 43.3 12.7 3.6 1.4
Controls are from the Norwegians pharmaceutical data base (NorPD) and between ages 41-80 in 2010 and alive by the end of 2010 Prospective deaths include all those between ages 41-80 who died between January 2010 and October 2010
to death than did the older age groups in the last few months (OR 1.8) for the oldest.
Discussion A considerable number of studies, including the present one, found an association between BZZ use and increased mortality [1–8]. The question is whether this association can be considered causal or whether an unadjusted confounder is Table 2 Odds ratio (OR) for BZZ use and for opioid use, both adjusted for age and sex Variable
Reference
a. BZZ use and mortality Sex Females Age in 10 years Age 41–50 BZZ usea No BZZ b. Opioid use and mortality Sex Females Age in 10 years Age 41–50 Opioid useb No opioids
OR
Confidence interval
0.47 2.20 2.30
0.45 2.16 2.20
0.49 2.25 2.40
0.54 2.36 1.62
0.52 2.31 1.56
0.56 2.41 1.69
a
BZZ use at any time over the 2-year exposure period prior to death/index period
b Opioid use at any time over the 2-year exposure period prior to the index/death date
present. The present study illustrated that the frequency of filling BZZ prescriptions increased with approaching death, culminating with the greatest frequency in the last few months before death. This pattern of gradually increasing BZZ accessibility until time of death is consistent with a gradual increase in symptomatic treatment of the increasing pain and other discomforts of approaching death. This pattern of BZZ availability would show itself as an association between BZZ use and increasing mortality, and may well be the confounder underlying the association found in this and many other studies. Opioid use which has not been implicated in mortality in the same way as BZZ has been was shown to have a similar association with mortality. Thus, we conclude that the association between BZZ availability and mortality can be at least be partially, and possibly be wholly, explained by an increase in symptomatic treatment of increasing discomfort as death draws near. As with all studies, this study has strengths and limitations. Strengths include the availability of the data for the entire Norwegian population, as well as confidence in the reliability of the data. Strengths also include the information available on each prescription, such as prescriber id, age, sex, ATC code, and date of death. One concern is the lack of information on the institutionalized since their medications could not be recorded in the NorPD. This is also the reason for the age cutoff at age 80 since institutionalization increases with age. In addition, many people are hospitalized at least some of the
640
Eur J Clin Pharmacol (2015) 71:637–642
Fig. 1 Percentage of population using BZZ prior to death/index date and age. Controls were derived from the Norwegians pharmaceutical data base (NorPD) and were between ages 41–80 in 2010 and alive by the end of 2010. Prospective deaths include all those in the NorPD who were between ages 41 and 80 in 2010 and who died between January 2010 and October 2010
time shortly before death, and one can assume that the persons in the death cohort were more likely to be hospitalized than the control group. Thus, they likely received more BZZ than is recorded. As a result, the increased BZZ use in the death cohort as indicated in this study may be considered an underestimate. Another potential ambiguity is the day of death within the 2-month death period. The 2-year exposure period was calculated backward from the 2-month period during which the date of death or the index date occurred. Thus, for deaths, there is a period of between a minimum of 1 day and a maximum of 60 days prior to the day of death for which no exposure was measured for this study. The advantage of this lag period is that some of the irregularities of the last few weeks of terminal illness, such as the increased potential for institutionalization or palliative sedation, are less likely to affect the data. A disadvantage of this lag period is that the number of days between beginning of the period and death is variable. For roughly one quarter of deaths, the lag period will be less than 2 weeks, but for about three quarters of deaths, the lag period will be more than 2 weeks. On average, we can expect this lag time to reduce the irregularities in BZZ use of the last while before death and, on average, we can expect this lag in Fig. 2 Odds ratios (OR) for BZZ use by months before index/death period adjusted for age and sex. a For all ages and both BZZ and opioids. b For BZZ stratified by age groups. See legend of Fig. 1 for description of controls and prospective deaths. Logistic regression models prospective deaths and controls were calculated for each point separately and adjusted for sex and age in single years even within the larger age group
exposure time to reduce the association between BZZ use and mortality, making the results of this study a more conservative estimate. Based on the pattern of BZZ use in the months before death, we concluded that the pattern of BZZ prescribing in the death cohort is consistent with treatment of increasing symptoms of pain and discomfort in the final months of life. It should not be a surprise that more of these types of medications are used in the last year of life [17, 18]. This pattern of use is most likely enough to explain the overall association between BZZ use and mortality found in this study and also that found in other studies. Others who found an association between BZZ and increased mortality have come to different conclusions. Kripke et al. concluded that the association was most likely causal, especially for cancer [1, 9, 13]. Major research projects were carried out, adjusting for many variables such as age, sex, smoking, BMI, marital status, health status, ethnicity, alcohol, and prior cancer, in order to adjust for potential confounding. Based on the positive association of BZZ and mortality remaining after adjusting for many potential confounding, Kripke et al. concluded that BZZ are dangerous drugs [9]. Although numerous variables were measured and adjusted
a) For all ages and both BZZ and opioids.
b) For BZZ stratified by age groups
Eur J Clin Pharmacol (2015) 71:637–642
for, what could neither be measured nor adjusted for were the increasingly severe symptoms of pain and other discomfort that often herald impending death, since information on severity of symptoms, other than that relating to number of conditions from which a person suffers, is rarely available. This variable then has become a confounder. A variety of other evidence also supports confounding rather than causality. One factor is the lack of specificity in the cause of death showing the association with BZZ, ranging from all-cause mortality to cancer to cardiovascular disease, to kidney disease, etc. [1–7, 9]. It is difficult to think of a biological reason for increased risk of mortality that is true for such a wide range of conditions. Other evidence for confounding also comes from similar patterns of increased BZZ with age and time to death. The present study shows that the youngest age group (41–50) shows a much larger association between BZZ and mortality than the oldest age group (71–80). Among published studies, those with an older study populations, e.g. over 65, show little association between BBZ and mortality [6, 12, 19, 20], while those with study populations with people under 60 are more likely to show an association [1, 2, 21]. Another factor shown to determine the size of the association between BZZ and mortality in the studies cited is the amount of time between the recording of BZZ use and death. The present study shows already a 50 % increase in use of BZZ by 24 months before death. Many serious diseases leading to terminal illness have uncomfortable symptoms such as pain or difficulty sleeping already several years before death [22, 23]; thus, the gradual increase in BZZ use starting years before death is not surprising. The association of BZZ and mortality was shown to increase with proximity of death. Thus, such an association is more likely to occur and be higher, if at least some of the study population has only a short time left till death. Of the studies showing an association between BZZ use and mortality, Kripke et al.’s 2012 study shows one of the highest OR but it also had one of the shortest time lapses between the measurement of BZZ exposure and detection of death, i.e. an average of 2.5 years [1]. This short exposure time as well as the relatively young study population would be expected to lead to a large association if the increases of BZZ use are due to the increase in of symptoms as death approaches. Other evidence comes from the use of other medications under similar circumstances; particularly, other medications use to relieve symptoms of severe illnesses are likely to occur prior to death. While opioids are known to have the potential to hasten death in cases of severe pain and high tolerance, that is different from the way BZZ have been considered dangerous even when used when the person is still healthy [9]. Still, this study shows opioids to have a similar association with mortality as BZZ use. Other medications with greater use before death were antidepressants [24] and antipsychotics
641
[25] and an increase in various drugs of symptom management [18]. Surely, we cannot consider that all of these drugs are dangerous and to be avoided because they appear to have an association with increasing the risk of death. Confounding by indication or reason for prescription [26] or reverse causation [27, 28] seems to be more plausible. In conclusion, the pattern of increasing BZZ use before death is consistent with its use as treatment of worsening symptoms with approaching death. Much evidence has been presented here supporting the notion that the association between BZZ and mortality is at least partially due to confounding, rather than being instrumental in initiating new disease leading to mortality. Whether the association between BZZ use and mortality is considered causal or confounding is an important distinction since the interpretation may determine how physicians use the medication. While some of Kripke’s message about the use of BZZ needs to be taken to heart [1, 9, 13, 29] and physicians need to be cautioned regarding the overuse or inappropriate use of BZZ, it is also true that when people are very uncomfortable when suffering from anxiety or lack of sleep, physicians should not avoid using BZZ for fear it increases the risk of death. Acknowledgments The authors would like to thank Drs. Svetlana Skurtveit, Christian Berg, and colleagues of the Norwegian Institute of Public Health, Oslo, Norway, for their support. Conflict of interest Neither author has any conflict of interest associated with any part of this study and article. Author involvement Both authors were involved at all stages of the research and writing of the article. While Dr. Neutel took the lead, Dr. Johansen was also involved in planning of the research methodology, the analysis of the data, and the editing of the paper.
References 1.
2. 3.
4.
5.
6.
Kripke DF, Langer RD, Kline LE (2012) Hypnotics' association with mortality or cancer: a matched cohort study. BMJ Open 2(1): e000850 Mallon L, Broman JE, Hetta J (2009) Is usage of hypnotics associated with mortality? Sleep Med 10(3):279–286 Belleville G (2010) Mortality hazard associated with anxiolytic and hypnotic drug use in the National Population Health Survey. Can J Psychiatry 55(9):558–567 Hausken AM, Skurtveit S, Tverdal A (2007) Use of anxiolytic or hypnotic drugs and total mortality in a general middle-aged population. Pharmacoepidemiol Drug Saf 16(8):913–918 Mallon L, Broman JE, Hetta J (2002) Sleep complaints predict coronary artery disease mortality in males: a 12-year follow-up study of a middle-aged Swedish population. J Intern Med 251(3): 207–216 Merlo J et al (1996) Increased risk of ischaemic heart disease mortality in elderly men using anxiolytics-hypnotics and analgesics. Results of the 10-year follow-up of the prospective population study "Men born in 1914", Malmo, Sweden. Eur J Clin Pharmacol 49(4):261–265
642 7.
8.
9. 10. 11.
12. 13.
14.
15.
16.
17.
18.
Eur J Clin Pharmacol (2015) 71:637–642 Fukuhara S et al (2006) Symptoms of depression, prescription of benzodiazepines, and the risk of death in hemodialysis patients in Japan. Kidney Int 70(10):1866–1872 Weich S et al (2014) Effect of anxiolytic and hypnotic drug prescriptions on mortality hazards: retrospective cohort study. BMJ 348:g1996 Kripke DF (2009) Do hypnotics cause death and cancer? The burden of proof. Sleep Med 10(3):275–276 Phillips B, Mannino DM (2005) Does insomnia kill? Sleep 28(8): 965–971 Ahmad R, Bath PA (2005) Identification of risk factors for 15-year mortality among community-dwelling older people using Cox regression and a genetic algorithm. J Gerontol A Biol Sci Med Sci 60(8):1052–1058 Jaussent I et al (2013) Hypnotics and mortality in an elderly general population: a 12-year prospective study. BMC Med 11:212 Kripke DF, Langer RD, Kline LE (2012) Do no harm: not even to some degree. J Clin Sleep Med JCSM Off Publ Am Acad Sleep Med 8(4):353–354 Furu K (2001) Drug utilisation in a public health perspective: Establishing a national prescription register in Norway. Nor J Epidemiol 11:55–60 Furu K et al (2010) The Nordic countries as a cohort for pharmacoepidemiological research. Basic Clin Pharmacol Toxicol 106(2):86–94 WHO, Collaborating Centre for Drugs Statistics Methodology: ATC/DDD Index 2011: Oslo, Norway, available from: http:// www.whocc.no last accessed 10/05/2013 Fahlman C et al (2006) Prescription drug spending for Medicare + Choice beneficiaries in the last year of life. J Palliat Med 9(4):884– 893 Langton JM (2013) Prescription drug use in the last 12 months of life: observations in a cohort of elderly patients, in International
Conference of Pharmacoepidemiology. Pharmacoepidemiol Drug Saf Montr. Abstract 703 19. Chen HC, Su TP, Chou P (2013) A nine-year follow-up study of sleep patterns and mortality in community-dwelling older adults in Taiwan. Sleep 36(8):1187–1198 20. Hays JC, Blazer DG, Foley DJ (1996) Risk of napping: excessive daytime sleepiness and mortality in an older community population. J Am Geriatr Soc 44(6):693–698 21. Kojima M et al (2000) Sleep patterns and total mortality: a 12-year follow-up study in Japan. J Epidemiol 10(2):87–93 22. Coleman MP et al (2008) Cancer survival in five continents: a worldwide population-based study (CONCORD). Lancet Oncol 9(8):730–756 23. Verdecchia A et al (2008) Patient survival for all cancers combined as indicator of cancer control in Europe. Eur J Public Health 18(5): 527–532 24. Brelin S et al (2013) Antidepressants to cancer patients during the last year of life— population-based study. Psychooncology 22(3): 506–514 25. Wang PS et al (2005) Risk of death in elderly users of conventional vs. atypical antipsychotic medications. N Engl J Med 353(22): 2335–2341 26. Strom B (2005) Pharmacoepidemiology. John Wiley & Sons Ltd., Chichester 27. Pottegard A et al (2013) Use of benzodiazepines or benzodiazepine related drugs and the risk of cancer: a population-based case-control study. Br J Clin Pharmacol 75(5):1356–1364 28. Kao CH et al (2012) Relationship of zolpidem and cancer risk: a Taiwanese population-based cohort study. Mayo Clin Proc 87(5): 430–436 29. Kripke DF (2007) Greater incidence of depression with hypnotic use than with placebo. BMC Psychiatry 7:42
