How to quantify deaths averted derived from interrupted time-series analyses

TLDR: In this paper, the authors compare the results of two different methodological approaches to quantify deaths averted using different two standard populations, and compare the impact of a) using interrupted time series (ITS) methodology vs fitting the trend before the intervention to predict the following 12 months and comparing the predicted monthly estimates of deaths with the actual numbers; and b) adjusting the time series either using the World Health Organization standard or the age distribution of Lithuania in the month before an excise tax increase.

Abstract: Background Interrupted time series (ITS) are an important tool for determining whether alcohol control policies, as well as other policy interventions, are successful over and above secular trends or chance Subsequent to estimating whether a policy has had an effect, quantifying the key outcomes, such as the number of prevented deaths, is of primary practical importance The current paper compares the results of two different methodological approaches to quantify deaths averted using different two standard populations Methods Time series methodologies were used to estimate the effect size in deaths averted of a substantial increase in excise taxation in Lithuania in 2017 We compare the impact of a) using ITS methodology vs fitting the trend before the intervention to predict the following 12 months and comparing the predicted monthly estimates of deaths with the actual numbers; and b) adjusting the time series either using the World Health Organization standard or the age distribution of Lithuania in the month before the intervention The effect was estimated by sex Results The increase in excise taxation was associated with a substantial decrease in all-cause mortality in all models considered ITS methodology and using the age-distribution of Lithuania were consistently associated with higher estimates of deaths averted Although confidence and prediction intervals were highly overlapping, the point estimates differed substantially The taxation increase was associated with 1,155 deaths averted in the year following the intervention (95% prediction interval: 729, 1,582), corresponding to 280% of all deaths in Lithuania in the respective year, for the model selected as best for planning policy interventions in Lithuania Conclusions Fitting a time series model for the time until the intervention, and then comparing the predicted time points with the actual mortality, standardizing to country-specific weights, was chosen as the best way to derive practically relevant effect sizes

Full text

1
How to quantify deaths averted derived from interrupted time-series
1
analyses
2
Huan Jiang, PhD
1,2
, Alexander Tran, PhD
1
, Gerhard Gmel, PhD
1,3,4
, Shannon Lange, PhD
1,5
, 3
Jakob Manthey, PhD
6-8
, Robin Room, PhD
9,10
, Pol Rovira, MSc
11
, Mindaugas Štelemėkas, PhD 4
12,13
, Tadas Telksnys, PhD
12
, Jürgen Rehm, PhD
1-2, 5-6, 11, 15-17
5
6
1.
Institute for Mental Health Policy Research, Centre for Addiction and Mental Health, 33 7
Ursula Franklin Street, Toronto, Ontario, Canada, M5S 2S1 8
2.
Dalla Lana School of Public Health, University of Toronto, 155 College Street, Toronto, ON, 9
M5T 1P8, Canada 10
3.
Addiction Medicine, Department of Psychiatry, Lausanne University Hospital and University 11
of Lausanne, Lausanne, Switzerland 12
4.
Department of Health and Social Sciences, University of the West of England, Bristol, UK 13
5.
Campbell Family Mental Health Research Institute, Centre for Addiction and Mental Health, 14
33 Ursula Franklin Street, Toronto, Ontario, Canada, M5T 2S1 15
6.
Institute of Clinical Psychology and Psychotherapy & Center of Clinical Epidemiology and 16
Longitudinal Studies (CELOS), Technische Universität Dresden, Chemnitzer Str. 46, 01187 17
Dresden, Germany 18
7.
Center for Interdisciplinary Addiction Research (ZIS), Department of Psychiatry and 19
Psychotherapy, University Medical Center Hamburg-Eppendorf (UKE), Martinistraße 52, 20
20246 Hamburg, Germany 21
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted March 26, 2021. ; https://doi.org/10.1101/2021.03.23.21254181doi: medRxiv preprint
NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.

2
8.
Department of Psychiatry, Medical Faculty, University of Leipzig, Semmelweisstraße 10, 22
04103 Leipzig, Germany 23
9.
Centre for Alcohol Policy Research, Building NR-1, La Trobe University, Plenty Rd. x 24
Kingsbury Rd., Bundoora, Victoria 3086, Australia 25
10.
Centre for Social Research on Alcohol and Drugs, Department of Public Health Sciences, 26
Stockholm University, 3rd floor, Sveavägen 160, 113 46 Stockholm, Sweden. 27
11.
Program on Substance Abuse, Public Health Agency of Catalonia, 81-95 Roc Boronat St., 28
08005, Barcelona, Spain 29
12.
Health Research Institute, Faculty of Public Health, Lithuanian University of Health Sciences, 30
Tilžės str. 18, 47181 Kaunas, Lithuania 31
13.
Department of Preventive Medicine, Faculty of Public Health, Lithuanian University of 32
Health Sciences, Tilžės str. 18, 47181 Kaunas, Lithuania 33
14.
World Health Organization / Pan American Health Organization Collaborating Centre, 34
Centre for Addiction and Mental Health, 33 Ursula Franklin Street, Toronto, Ontario, 35
Canada, M5S 2S1 36
15.
Faculty of Medicine, Institute of Medical Science, University of Toronto, Medical Sciences 37
Building, 1 King’s College Circle, Room 2374, Toronto, Ontario, Canada, M5S 1A8 38
16.
Department of Psychiatry, University of Toronto, 250 College Street, 8th floor, Toronto, 39
Ontario, Canada, M5T 1R8 40
17.
Department of International Health Projects, Institute for Leadership and Health 41
Management, I.M. Sechenov First Moscow State Medical University, Trubetskaya str., 8, b. 42
2, 119992, Moscow, Russian Federation 43
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted March 26, 2021. ; https://doi.org/10.1101/2021.03.23.21254181doi: medRxiv preprint

3
44
JR acknowledges funding from the Canadian Institutes of Health Research, Institute of 45
Neurosciences, and Mental Health and Addiction (CRISM Ontario Node grant no. SMN-13950). 46
Research reported in this publication was also supported by the National Institute on Alcohol 47
Abuse and Alcoholism of the National Institutes of Health (NIAAA) [Award Number 48
1R01AA028224-01]. 49
50
Corresponding author: H. Jiang 51
Address: Centre for Addiction and Mental Health, 33 Ursula Franklin Street, Room T420, 52
Toronto, Ontario, Canada, M5S 2S1; Tel: + 416-535-8501 Ext 36173 53
Email: hedy.jiang@utoronto.ca 54
55
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted March 26, 2021. ; https://doi.org/10.1101/2021.03.23.21254181doi: medRxiv preprint

4
Abstract
56
Background 57
Interrupted time series (ITS) are an important tool for determining whether alcohol control 58
policies, as well as other policy interventions, are successful over and above secular trends or 59
chance. Subsequent to estimating whether a policy has had an effect, quantifying the key 60
outcomes, such as the number of prevented deaths, is of primary practical importance. The 61
current paper compares the results of two different methodological approaches to quantify 62
deaths averted using different two standard populations. 63
Methods 64
Time series methodologies were used to estimate the effect size in deaths averted of a 65
substantial increase in excise taxation in Lithuania in 2017. We compare the impact of a) using 66
ITS methodology vs. fitting the trend before the intervention to predict the following 12 months 67
and comparing the predicted monthly estimates of deaths with the actual numbers; and b) 68
adjusting the time series either using the World Health Organization standard or the age 69
distribution of Lithuania in the month before the intervention. The effect was estimated by sex. 70
Results 71
The increase in excise taxation was associated with a substantial decrease in all-cause mortality 72
in all models considered. ITS methodology and using the age-distribution of Lithuania were 73
consistently associated with higher estimates of deaths averted. Although confidence and 74
prediction intervals were highly overlapping, the point estimates differed substantially. The 75
taxation increase was associated with 1,155 deaths averted in the year following the 76
intervention (95% prediction interval: 729, 1,582), corresponding to 2.80% of all deaths in 77
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted March 26, 2021. ; https://doi.org/10.1101/2021.03.23.21254181doi: medRxiv preprint

5
Lithuania in the respective year, for the model selected as best for planning policy interventions 78
in Lithuania.
79
Conclusions 80
Fitting a time series model for the time until the intervention, and then comparing the 81
predicted time points with the actual mortality, standardizing to country-specific weights, was 82
chosen as the best way to derive practically relevant effect sizes. 83
Keywords or phrases (max 5): 84
Alcohol – all-cause mortality – taxation – interrupted time series – effect size
85
86
. CC-BY-NC-ND 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)
The copyright holder for this preprint this version posted March 26, 2021. ; https://doi.org/10.1101/2021.03.23.21254181doi: medRxiv preprint

Frequently asked questions

Q1. What contributions have the authors mentioned in the paper "How to quantify deaths averted derived from interrupted time-series analyses" ?

Jiang et al. this paper proposed a framework for mental health policy research based on the work of the Centre for Addiction and Mental Health. 

Q2. What is the reason for the higher estimated number of deaths averted by the classic ITS?

The higher estimated number of deaths averted by 295 the classic ITS model may be the result of the modelling strategy, which assumed the same 296effect size for the remainder of time period, i.e., for longer than one year. 

Q3. What is the way to compare the mortality of a high-income country?

Based on their best estimates using the 318Lithuanian age distribution, reducing the overall adult mortality in a high-income country like 319 Lithuania by 4.83% for men, 0.84% for women and 2.80% overall is a great achievement for a 320 single intervention, and truly deserves the label of a “best buy”. 

Q4. What is the effect of taxation on all-cause mortality?

264 However, when applied correctly, alcohol taxation constitutes a powerful tool that can result in 265 sizable all-cause mortality reductions. 

Q5. Who supported the research in this article?

325 Research reported in this publication was also supported by the National Institute on Alcohol 326 Abuse and Alcoholism of the National Institutes of Health (NIAAA) [Award Number 327 1R01AA028224-01].