Trends in Noticing and Responding to, Health Warning Labels on Cigarette Packages Among Adults Who Smoke: Findings from the ITC Four Country Surveys Between 2002 and 2022

Abstract

Introduction

We present data on 20-year (2002–2022) trends in reactions of smokers to Health Warning Label (HWL) changes (eg, increasing warning size, requiring graphic images, mandating standardized packaging) in Canada, England, and Australia, compared to the US where HWLs did not change.

Methods

We analyzed weighted data from 16 waves of the International Tobacco Control Four Country Survey, comprising 99 438 observations from 49 034 adults (ages 18+) who smoked cigarettes (daily or non-daily), tracking indicators of HWL effectiveness including noticing HWLs and related thoughts about harms from smoking and quitting.

Results

The first HWL change studied in each country (Canada-2001; England-2003; Australia-2006) had the largest impact on all indicators of HWL effectiveness; subsequent changes in all three countries had less or no additional impact. Over 20 years, noticing and quitting thoughts (for daily smoking) decreased significantly in the US. In the last three waves (2018–2022), noticing remained lowest in the US. Thinking about harm and quitting were lower in the US than in Canada and England, but in Australia, levels of “Thinking about harms” were significantly lower than in the US. People smoking non-daily maintained higher levels on some key measures than those smoking daily.

Conclusions

In countries with long histories of public education about smoking harms, periodically strengthening HWLs on cigarette packs appears to have diminishing impacts on smokers’ reactions with long-term impacts of strong warnings small, especially for people smoking daily. The findings suggest short-term impacts of HWLs relate to the change from previous HWLs and not their absolute magnitude.

Implications

HWLs on cigarette packs remain important for discouraging smoking, but the effects of strengthening warnings on smokers’ reactions decline with time and are smaller where HWLs were already strong. Sustained effects may be greater for non-daily smokers. Countries with weak or no HWLs may benefit in the medium term from stronger HWLs, though returns diminish as HWL coverage increases. Future smoking models should not assume long-term large cessation effects from stronger HWLs as they can do in the short term but allow for a range of smaller impacts. This study did not assess HWLs’ impact in preventing uptake.

Introduction

Cigarette smoking is the leading preventable cause of morbidity and mortality causing the deaths of over 8 million people globally each year.^1,2 Yet, ⁓20% of adults smoke cigarettes worldwide.¹ Strong health warning labels (HWLs) on cigarette packs is an accessible and cost-effective way to educate people about the harms and thus discourage use. For example, smoking 20 cigarettes per day generates over 7000 interactions with a pack annually. WHO Framework Convention on Tobacco Control (FCTC) guidelines recommend rotating pack warnings, covering over 50% of the principal display areas and including pictures.³

The US first implemented text-only HWLs in 1966 and last modified them in 1985 with one of four rotating text-only warnings on one side of the pack. This provides a natural control condition for studying the impacts of more recent HWL changes in other countries in more recent years. We consider the effects on smokers of HWL changes on cigarette packs since 2001 when Canada implemented full-color pictorial HWLs on half the packs’ front⁴ ⁓16 months before the International Tobacco Control (ITC) Project commenced data collection. Pictorial warnings on ˂50% of the front were adopted by Australia in 2006 and UK/England in 2009. Australia increased the size to 75% of the front in 2012 while mandating standardized or plain packaging. Other notable changes were, UK/England increased the HWLs to 65% in 2016, mandated standardized packaging from May 2017, and changed only warning content in 2021. Canada increased the warning size from 50% to 75% of the front in 2012 and introduced standardized packaging in 2020 (see Supplementary Figure S1). All three other countries⁵ had HWLs well before the study period, meaning the study is about the effect of strengthening HWLs.

Early findings on population-level effects of small HWLs demonstrated knowledge gain and reactions toward quitting, such as thinking about smoking harms, thinking about quitting or forgoing cigarettes.^6,7 Some of these effects were observed in experimental studies, showing a possible causal pathway.⁸ Longitudinal studies have also shown that warning reactions predict subsequent quit attempts.^7,9–11 Several papers have examined aspects of the data presented here. In summary, the first three warning changes evaluated (Canada’s 2001 pictorial warnings, UK/England’s 2002 warning panel, and Australia’s 2006 pictorial warnings) all substantially impacted all or nearly all the measures considered here¹² but impact declined at similar rates^12,13 with no wear-out reduction observed from annual warning rotation in Australia.¹² Subsequent changes have been less studied. There were strong positive impacts of standardized packaging in Australia (accompanied by increasing HWL size),¹⁴ albeit smaller than the effect of previous changes; a small overall impact when introduced in UK/England in 2017, again with some HWL strengthening, but no clear evidence of impacts measures when introduced in Canada in 2020 without changing the HWLs.¹⁵

The broader literature confirms that HWL impact depends on many factors, including size, use of graphics, placement (front face is best); and contrast with non-warning elements, message content, and novelty.^16–22 Noticing HWLs is associated with increased thinking about harms and increased reported avoidance²³ and can lead to increased quit attempts.^9,16,24,25 Concerns about reactance leading to avoidance or inhibiting thoughts of concern, reducing quitting²⁶ although demonstrated at an event-by-event level, do not appear to reduce quitting interest over time.^9,27–29 Indeed over extended periods, avoidance is associated with greater thinking about the harms.^9,11 While we previously found evidence of reduced impacts of HWLs over time (wear out) (eg,¹²), the introduction of strong warnings in Thailand showed sustained impacts over 2 years.,³⁰ so we should not assume that wear-out always occurs, at least as rapidly as the published research suggests. Effects might be more likely to persist in countries without strong histories of anti-smoking communications through other media.

This paper examines the impacts of HWL changes on smokers over 20 years (2002-2022) in three countries–Canada, England, and Australia, compared to the US where no changes occurred. It encompasses many findings previously reported^12,16,19 but extends these to more comprehensively explore: (a) reactions to HWLs in relation to levels and changes in the size of the warnings, the use of graphic images and standardized packaging, whether reactions to HWLs are proportional to warning size and strength, and if potency decreases over time; (b) overall trends in reactions to HWLs; and (c) whether there are recent country differences in reactions to HWLs, particularly in comparison with the US. Analyses presented herein also explore (d) any differences as a function of daily versus non-daily smoking. HWLs are potentially most impactful for people with less dependence.³¹

Methods

Sample and Design

Data comes from the ITC Survey, a cohort study of adults (aged 18+ years) which replenishes the sample using the same sampling frame as for recruitment.³² Data reported here are from participants from Canada, the US, England, and Australia who smoked at least monthly (daily or non-daily) at survey time(s). The dataset includes 16 survey waves between 2002 and 2022, with between 13 and 15 surveys in each country with inter-wave intervals up to ⁓2 years (see Supplementary Table S1 for details). The first survey (October to December 2002) was immediately after an increase in the size of HWL in the UK was mandated, but before these new packs were widely available, and ⁓16 months after the introduction of Canada’s pictorial HWLs.

The samples were originally recruited using stratified probability sampling methods using random digit dialing (RDD) in each country from 2002 until 2007. Starting from Wave 7 (2008), a web-based survey was introduced in addition to RDD with only web-based surveying since 2016 (Wave 11). New recruitment used commercial panels from 2013 (Wave 9) Complete details of the study are published in technical reports (https://itcproject.org/methods/technical-reports/).

We analyzed data of 49 034 unique respondents (12 580 from Canada, 12 824 from the US, 14 920 from England, and 8710 from Australia) who participated in at least one wave, including 99 438 observations for analysis; (87 072 daily; 12 366 non-daily). Table 1 shows the sample characteristics.

Table 1.

Sample characteristics of participants in the survey waves (n = 99 438)

Variable	Overall n = 99 438 (weighted %)	US n = 25 016 weighted %	Canada n = 25 992 weighted %	England n = 26 251 weighted %	Australia n = 22 179 weighted %
Sex
Male	46 359 (53.7)	54.2	53.9	52.2	54.5
Female	53 079 (46.3)	45.8	46.1	47.8	45.5
Current age
18–24	11 013 (9.8)	9.5	8.7	11.9	8.9
25–39	23 442 (31.3)	29.3	28.2	33.3	34.7
40–54	33 044 (32.8)	33.8	35.6	28.4	33.6
55–max	31 939 (26.1)	27.4	27.5	26.4	22.8
Education
Low	42 423 (43.7)	46.6	39.1	34.8	56.2
Moderate	35 800 (38.6)	37.8	40.8	46.8	27.1
High	20 625 (17.1)	15.5	19.7	16.6	16.3
No answer	588 (0.7)	0.1	0.4	1.8	0.4
Income
Low	29 870 (28.8)	36.9	28.1	24.8	25.6
Moderate	31 765 (32.4)	31.7	31.7	36.4	29.1
High	31 276 (32.1)	27.3	32.9	30.5	38.7
No answer	6465 (6.6)	4.2	7.2	8.3	6.6
Cigarettes per day
Non-daily	12 366 (11.3)	10.6	13.0	12.5	8.7
1–10 Daily	27 558 (28.4)	30.3	27.4	30.8	24.7
11–20 Daily	40 391 (41.5)	41.5	38.9	44.4	41.0
21–30 Daily	13 567 (13.7)	10.7	16.7	8.6	19.5
≥31 Daily	4385 (4.4)	5.9	3.6	2.6	5.7
No answer	869 (0.7)	0.9	0.4	1.0	0.4

Note: Frequencies are based on unweighted data.

HWL Impact Measures

Table 2 describes the six HWL impact measures with the years the questions were asked and the binary response coding used for the analysis. The first column uses the summary term we use hereon. Noticing is a prerequisite to impact. Think harms is the proximal intended impact and Think quit is the desired subsequent cognition if it is to impact positively on smokers’ choices. These are the measures we focus on. The other three measures are of possibly less importance for understanding HWL impacts. Notice first was a measure first introduced in 2011 to assess the impacts of standardized packaging. Avoidance is an immediate impact measure but has not been shown to predict intentions independently, while Forgo is an immediate behavioral response to warnings, and while associated with quitting intentions, but is negatively associated with quit success, so it may also assess difficulty in quitting.

Table 2.

The six health warning label impact measures along with the years the questions were asked and the binary response coding used for the analysis

HWL impact variable	Question	Response coding (options)
Noticing (all waves)	In the last month, how often have you noticed the warning labels on cigarette packages?	(1) Very often, Often (0) Sometimes, Rarely, Never, Don’t Know
Noticed firsta (from 2011)	What do you usually notice first…	(1) First: HWLs (0) Other aspects such as branding, Don’t Know
Avoiding (2002–2006)	In the last month, have you made any effort to avoid looking at or thinking about the warning labels: By, (1) covering the warnings up; (2) by keeping the pack out of sight; (3) by using a cigarette case or some other pack; and (4) by not buying packs with particular labels	(1) Avoiding: Yes to any of the 4 (0) No or Don’t Know to all 4
Avoiding (2007–2022)	In the last month, you made an effort to avoid looking or thinking about warning labels	(1) Many times, A few times (0) Once, Never, Don’t know
Think harm (2003–2022)	To what extent, if at all, do the warning labels made you think about the health risks of smoking	(1) A lot, Somewhat (0)A little, Not at all, Don’t know
Think quit (2003–2022)	To what extent, if at all, do the warning labels on cigarette packs make you more likely to quit smoking?	(1) A lot, Somewhat (0) A little, Not at all, Don’t know
Forgo (2003–2022)	In the last month, have the warning labels stopped you from having a cigarette when you were about to smoke one?	(1) Many times, A few times (0) Once, Never, Don’t Know

^aQuestion was first asked in 2011 (and only in Australia) before being asked consistently since 2012.

Statistical Analysis

We provide the proportions of respondents for each outcome (eg, Think-Harm “somewhat” or “a lot”). The proportions were weighted by country to accurately represent the age-sex strata estimates in the population and account for non-response and survey design.³³ No other adjustments were made. Most of the results are presented in graphs to show the impact of changes over time. Non-overlapping 95% confidence intervals (CIs) between data points indicate a statistically significant difference. If two values have the same 95% CI value for a lower and an upper estimate, the probability that the two mean values differ is p=.000625. For parsimony, we only provide statistics (95% CIs) for selected cases where effects may not be clear (see Supplementary materials for greater detail).

Stata/MP 18 (StataCorp, 2023, College Station, TX, US) was used for analysis. Specific comparisons were made, where possible, of the absolute magnitude of responses in the wave following a change in warnings or packaging and the magnitude of the increase from the wave before to the wave following the change. Sensitivity analyses restricted to new recruits at each wave found no notable effects of retaining participants (data available on request). Trend analysis in each of the four countries was done using panel logistic regression (only done for the five key measures assessed from the early waves of data). The formal analysis focussed on linear trends as non-linearities are at least partly attributed to specific intervention effects. Finally, we tested for country differences in HWL reactions in the last three waves using panel logistic regression to assess the current situation (analysis of the last wave is in Supplementary Table S2 but has less power to find differences). For daily vs non-daily smoking, comparisons were made on trends in the US and on differences in the last three waves by country for all relevant measures.

Results

We focus on three of the measures: noticing, think harm and think quit (see Figure 1), but consider the other three measures, especially where they respond differently (see Supplementary Figures S2–S4). The estimates in the figures for reactions in each country and wave are in a timeline by the month when half the sample was surveyed (see Supplementary Table S3) and are marked with HWL change implementation dates (vertical dotted lines), denoted as Ca# (eg, Ca1, Ca2, Ca3), Au# and En# for Canada, Australia, and England, respectively. Mandating of changes typically occurs several months before revised HWLs appear on packs, so when implementation was closely followed by a survey, the second survey post-implementation was used to identify effects.

Figure 1.

Respondents reporting (A) noticing (B) think harm, and (C) think quit in the past month between 2002 and 2022. Note: The vertical dotted lines (Ca#, Au#, En#) indicate the implementation of the new health warnings policies in the respective countries (eg, Ca1 refers to implementation of pictorial health warning labels in Canada in 2001). The x-axis represents the median date for each survey- wave for the four countries. ^*questions were asked only from wave 2 onwards.

As seen in Figure 1, reactions to the US HWLs were nearly always lower than those in the other countries, which all had stronger HWLs from the commencement of the study period. Responses significantly varied in the US on all measures although no HWL changes occurred. There were also shifts in other countries in waves quite separate from HWL changes. Major ones are noted below.

The greatest variation in the measures was in the three countries where HWL changes occurred in earlier waves (up to ⁓2006) where levels of key measures were high in the years following HWL changes, but less or undetectable to HWL changes that occurred in subsequent years. These patterns are described in more detail below. Table 3 presents details on the size of changes in responses in the peak wave following a HWL change. Where a similar pattern was found in the three countries with stronger and/or changed warnings, we describe these as the “strong warning countries” to contrast them with the US.

Table 3.

Changes in proportion^a responding from the wave immediately before to the first wave after health warning label changes

	Noticing	Avoiding	Think harm	Think quit	Forgo
England 2002 (En1)	0.40 (0.37–0.43)	–	–	–	–
Australia 2006 (Au1)	0.28 (0.24–0.32)	–	0.18 (0.14–0.22)	0.14 (0.11–0.17)	0.08 (0.06–0.10)
England 2009 (En2)	–0.02 (–0.08 to 0.04)	0.02 (–0.03 to 0.07)	0.01 (–0.04 to 0.06)	–0.01 (–0.06 to 0.04)	0.01 (–0.02 to 0.04)
Canada 2012 (Ca2)	0.08 (0.03–0.13)	0.09 (0.05–0.13)	0.09 (0.04–0.14)	0.07 (0.03–0.11)	0.01 (–0.02 to 0.04)
Australia 2012 (Au2)	0.06 (0.00–0.12)	0.19 (0.14–0.24)	0.03 (–0.03 to 0.09)	0.03 (–0.02 to 0.08)	0.02 (–0.02 to 0.06)
England 2016 (En3)	–0.09 (–0.14 to –0.04)	0.03 (0.00–0.06)	0.01 (–0.03 to 0.05)	0.00 (–0.03 to 0.03)	–0.04 (–0.07 to –0.01)
England 2017 (En4)	0.1 (0.07–0.13)	0.08 (0.06–0.10)	0.05 (0.02–0.08)	0.03 (0.01–0.05)	0.03 (0.01–0.05)
Canada 2020 (Ca3)	0.00 (–0.03 to 0.03)	–0.01 (–0.02 to 0.02)	0.01 (–0.02 to 0.04)	–0.01 (–0.03 to 0.01)	–0.01 (–0.03 to 0.01)

NB. Analyses of weighted data using the binary categorizations indicated in Table 2. The codes in brackets (eg, En1, Ca3) refer to the terms used to signal the changes in the Figures and text. Supplementary Table S4 presents analysis restricted to those smoking daily.

^aAbsolute change (95% CIs).

Noticing HWLs Often or Very Often (Noticing):  Figure 1 shows that in the stronger warning countries, Noticing HWLs “often” or “very often” peaked at or above 60% (0.6, as a proportion) in early waves, but was consistently ˂40% in recent waves. The US level remained ⁓30% except between 2009 and 2018 where it was significantly lower. The initial introduction of stronger HWLs in England (En1) and Australia (Au1) from text warnings covering less than half of the pack front led to a sharp increase in noticing, of 40% (0.40) and 28%, respectively (see Table 3), by far the largest changes observed. The highest level of noticing in Canada occurred at Wave 1 ⁓16 months after pictorial HWLs covering 50% of the pack front were introduced (Ca1). Subsequent changes in all countries either had lower peaks (Au2, En4, and Ca2) or no impact evident (En2, En3, and Ca3).

Think-Harms: The proportion of thinking about health harms due to HWLs ranged from ⁓25% to 50% (Figure 1), with reactions generally higher from 2002 to 2006 than in subsequent years. Levels in Canada were usually higher than in the other countries. The highest peak (50%) was in Australia following Au1. Reactions in England rose following En3 and En4 to the levels of Canada and remained higher than for the US and Australia thereafter, although still below the pre-2007 England levels. By contrast, following a brief rise associated with the 2012 changes, Australian reactions have declined markedly to below 30%.

Think-Quit: The proportion reporting that HWLs stimulated quit-related thoughts in the past month (Think-Quit) was lower overall than for Think-Harms, largely between 10% and 30% (Figure 1). All countries peaked in earlier waves (Canada in 2004, the US in 2003 and 2004, England in 2003, and Australia in 2006) with the largest peak following Au1. Subsequent HWL changes had mixed impacts (see Table 3). Think Quit rose between 2011 and 2013 in England with an upward trajectory in recent waves mirroring the pattern for Think-Harms.

Noticed HWLs before other pack elements  Noticed-First was first asked in an Australia-only wave in 2011, the wave before Au2 (see Supplementary Figure S2); it sharply increased in the wave following the change (30.3%–61.2%), followed by some decline (to 47.3% and to 39.8% in the subsequent two waves). There was also a marked increase in England following En4 (19.8%–45.6%) and relatively little decline thereafter. The proportion who Noticed first in Canada was relatively high (40%) when first measured in 2013, declining somewhat thereafter, with no increase following the 2020 change to standardized packaging. Levels in Australia, England and Canada remained above 30% from the point where they first rose above it, while levels in the US remained below 10% throughout.

Avoiding HWLs: Between 2002 and 2006, the proportion reporting any HWL avoidance peaked in Canada in 2002 (31.8%) after Ca1, in England in 2003 (20.1%) after En1 and in Australia in 2006 (36.1%) following Au1 but remained between 9.1% and 12.8% in the US (Supplementary Figure S3). Note that caution is needed when comparing these and later results because changing from four to one avoidance question in 2007 may have impacted results. The proportions who avoided HWLs between 2007 and 2022 peaked in Australia in 2013 (33.3 %) following Au2, in Canada in 2013 (21.9%) after Ca2 and in England in 2018 (25.9%) after En3 and En4, with the levels reported higher than the 2003 peak for this and the two subsequent waves. The proportion who avoided HWLs in the US remained low throughout this period (<10%).

Forgo: The proportion reporting forgoing cigarettes due to noticing the warnings remained low throughout (⁓10%) in all countries, and the change patterns were smaller, but largely consistent with those reported for the other measures (see Supplementary Figure S4).

Standardized Packaging: While standardized packaging in Australia (Au2) and England (En4) was associated with some increased reactions to HWLs (as noted above), in both cases, it was accompanied by increased warning size. In Canada, no changes in reactions to the unchanged HWLs were found following the introduction of standardized packaging.

Trends over Time: Examination of linear trends across the entire (or nearly entire) 20-year period (see Table 4) is most informative for the US as there were no changes to HWLs, so while the linear trends are similar in the other countries, they are more difficult to interpret given the large non-linearities created by the HWL changes. In the US, levels of Noticing, Think-Quit and Forgo declined over time (ie, ORs ˂1.0), while Avoiding (from 2007) increased. There was no significant change for Think-Harms. Generally, trends in the other three countries were similar, although Think harms significantly declined in all three countries.

Table 4.

Odds ratios for linear changes in levels of warning measures (2002–2022, daily or non-daily smoking)

Warning measure	US	Canada	England	Australia
	OR (95% CIs)	OR (95% CIs)	OR (95% CIs)	OR (95% CIs)
Noticing	0.967 (0.957–0.977)	0.911 (0.904–0.918))	0.903 (0.896–0.912)	0.937 (0.928–0.946)
Avoida	1.072 (1.040–1.104)	1.059 (1.040–1.080)	1.156 (1.129–1.184)	0.950 (0.928–0.973)
Think-harm	0.996 (0.985–1.006)	0.967 (0.958–0.975)	0.990 (0.980–1.0002)	0.966 (0.956–0.977)
Think-quit	0.958 (0.946–0.970)	0.964 (0.954–0.973)	0.986 (0.974–0.998)	0.967 (0.955–0.978)
Forgo	0.982 (0.968–0.996)	0.981 (0.970–0.993)	1.018 (1.004–1.031)	0.979 (0.964–0.995)

^aFrom 2007. Note: Refer to Supplementary Table S5 for trends separately for daily and non-daily smoking.

Recent Responses by Country: Next, we examined the recent waves focusing on the last three (see Table 5, Supplementary Table S2 provides comparable analyses for the last wave alone). Canada and England had higher responses than those in the US on all measures. By contrast, while respondents from Australia were more likely to report Noticing, Noticed-First and Avoiding, they were less likely to Think-Harm, and were not significantly different to the US on Think-Quit or Forgo.

Table 5.

Magnitude of differences of each of the three countries with strong warnings from the US reactions, waves 2018–2022 combined (daily or non-daily smoking)

Country	Noticinga	Noticed firsta	Avoidinga	Think harma	Think quita	Forgoa
US	Reference	Reference	Reference	Reference	Reference	Reference
Canada	1.64 (1.47–1.84)	6.40 (5.47–7.49)	2.70 (2.30–3.17)	1.38 (1.24–1.54)	1.40 (1.23–1.60)	1.49 (1.25–1.77)
England	1.90 (1.70–2.12)	9.95 (8.50–11.64)	3.97 (3.39–4.66)	1.25 (1.12–1.40)	1.48 (1.29–1.69)	1.67 (1.40–1.99)
Australia	1.36 (1.17–1.57)	7.24 (6.04–8.68)	2.24 (1.83–2.74)	0.81 (0.69–0.94)	0.94 (0.78–1.14)	0.83 (0.63–1.10)

^aOdds ratios (95% CIs in brackets).

Daily Versus Non-daily Smoking: Paralleling our analysis on trends, we tested for interactions with smoking frequency for linear trends for the US warnings and found no significant interaction with any measure (Supplementary Table S6). Overall, US non-daily smoking participants (compared to daily) were more likely to report Noticed-First (1.57; 1.20–2.06), Avoiding (1.255; 1.02–1.55), Think-Harm (1.42; 1.26–1.61), Think-Quit (1.59; 1.39–1.82), and Forgo (1.91; 1.63–2.25) but not Noticing.

We then looked at differences across the last three waves of data (since 2018). Across all countries there were (see Supplementary Table S7) no clear differences in responses by smoking frequency for Noticing, and Avoiding), but non-daily smoking participants were more likely to report Noticed-First (1.63; 1.20–2.23) (not analyzed for 20-year trends) and reported stronger reactions to Think-Harms (1.54; 1.25–1.89), Think-Quit (1.55; 1.18–2.03), and Forgo (2.24; 1.67–2.00), all larger effects than those for the entire period. There was some evidence of an interaction with country, at least in England, with people smoking non-daily relatively less likely to report Noticed-First and more likely to Think-Quit than the dailies.

Discussion

The findings of this study build on previous publications using some data included here on the population-level impacts of HWLs over 20 years in four countries with strong histories of public education around the harms of smoking.²⁵ We found that the self-reported impact of the first HWL change in each country in our study was larger than for later changes. We further confirmed that the increases in reactions following introduction of strengthened HWLs dissipate with time^13,16,19 and provide strong evidence that the measured impacts of strengthened warnings do not relate to their absolute magnitude, but to the relative change from the previous HWL. In the case of Australia, at least one quit-related indicator of impact, Think-Harms, is now lower than, and the other two are similar to, the much weaker US warnings. However, stronger reactions on all quit-related measures remain for Canada and England where HWL changes were more recent. In periods separated from HWL changes, differences between countries in reactions are small, as evidenced in the last three waves of data. The evidence, primarily from the US trends, suggests that the impacts of HWLs overall may have diminished over the 20-year period. Finally, we provide new evidence that non-daily smoking is associated with stronger quit-related reactions.

Our findings indicate that over 20 years, in all four countries, reactions to HWLs on most measures have declined from their peaks, including the most important ones of Think-Harms and Think-Quit, despite the HWLs markedly strengthening in three of the countries. The peak levels we identified did not include the addition of pictures for England, so size changes may be more important than pictures. The findings that subsequent reactions to HWL changes were lesser, even though the newer HWLs were made stronger, including adding pictures in England, suggests the increase in reactions is to the relative incremental change in the HWL, or even to the noticeability of the change, rather than the absolute size/strength of the resulting HWLs.

We add to previous findings,^14,15,23 particularly with the England data showing that standardized packaging appears to have measurable effects on HWLs when accompanied by strengthening of the HWLs, suggesting that the HWL change is the main cause of the impacts.³⁴ That said, we only have evidence on one country (Canada) where there was no accompanying change in HWL size, or implementation of standardized packaging with persisting weak HWLs, so warn against generalizing from our findings. We also note that the primary rationale for standardized packaging is to reduce smoking uptake among young people,¹⁵ something not examined in our study.

Previous research¹² showed a relatively even decline in warning impacts over ⁓5 years. The more rapid decline in impacts of the 2012 changes in Australia and possibly slower ones to the 2017 change in England, suggests there may be no common rate of wear-out. We have no clear ideas as to what factors might contribute to the observed differences.

The stronger reactions of non-daily smokers, to noticing and the quit-related measures are notable. It suggests a more persistent, albeit attenuated, positive effect among this group, which may relate to discouraging more experimental smoking, and is suggestive of a prevention effect for HWLs. The findings of stronger reactions among those smoking less have important methodological implications. They suggest that participants responded to our questions in terms of “given the opportunity” rather than making absolute estimates because someone smoking non-daily should see HWLs less frequently than someone smoking daily.

The above analysis suggests that, as US HWLs are less noticed, they may have less effect on quit-related measures than the similar estimates we found for the direct measures (eg, Think quit) would indicate because responses are conditioned on their lower level of noticing. If true, this would suggest a greater continuing benefit of the stronger warnings, and in the case of Australia, would likely mean that their HWLs continue to have more overall impact as is still the case in Canada and England, if we accept the US as a valid baseline control. We plan to explore the predictive effects of the quit-related measures, both as they are currently used and in combination with frequency of noticing to see if the combination improves predictive value for more proximal measures of quitting.

The most immediately plausible explanation of our findings overall is that in countries with a long experience of HWLs, they may have lost much of their potency among people who have established smoking habits, although a little less among people who smoke non-daily (less dependent). This would suggest that dependent use may lead to the warnings losing motivational potential, perhaps due to the difficulty of taking appropriate action (ie, quitting).

We now consider possible alternative explanations for our findings. First, methodological changes in sample recruitment and data collection³³ are unlikely as we saw no discontinuities around sampling and surveying changes. Furthermore, non-response and sample bias were accounted for by appropriately weighting the data. Could it be due to the measures not measuring what is critical for effectiveness? We assess consciously attributing reactions (eg, thinking about quitting), but these may not relate to unconscious motivating forces which might persist; forces if real, that would be unlikely to be detected using self-report methods. Alternatively, given that the average person only makes around two quit attempts per year, occasionally stimulating thinking about quitting may be as effective as more frequent responses, most of which fail to lead to action. This explanation might explain the overall trends, but not the narrowing of reactions between strong and weak warning regimes. That said, it will be useful to explore whether the predictive power of warning reactions for prospectively predicting quit attempts has changed, with any reduction consistent with reduced potency in this domain.

Over the study’s duration, smoking prevalence declined in all four countries, so the proportion of people resistant to quitting may have increased, leading to decreased responsiveness to HWLs. However, as this is the population of remaining adult smokers, it does suggest reduced potency of HWLs to contribute to further declines in prevalence.

Several statistically significant changes in reported reactions cannot be attributed to changes in HWLs. For example, changes in the US between 2007 and 2018 when reported Noticing of HWLs was appreciably lower as was reporting of Think harm and Think quit, albeit for shorter periods. Although the questions were asked directly relating to the HWLs, the pattern of responding suggests influences on responses from other sources. One possibility is that the importance of HWLs may be modulated by public education from other channels (eg, in social or mass media and point-of-sale warnings³⁵) or these sources may contribute to generating the responses. These factors work similarly to HWLs^36,37 and can complement them.³⁸ Therefore, HWLs may extend warnings in other media, and hence participants may not attribute their responses specifically to the HWLs. The US gave its Food and Drug Administration authority to regulate tobacco products in mid-2009 following a period of public debate, so an effect of this is plausible. That other sources of information about the harms might influence responding might also help explain the Australian data, which varied from other countries in several respects, most notably of changes attributable to HWLs declining to levels similar to, or in some cases below, US levels, a pattern not seen in either Canada or England. Australia has had a reputation for its strong anti-smoking campaigns, particularly focussing on adverse health effects but also regular increases in tobacco tax since 2010. It is possible that the low levels of responding overall in Australia could be due to HWLs playing a lesser role in prompting health concerns or the increased price becoming a more salient motivator.³⁹ A background of strong anti-smoking measures may concurrently heighten engagement with HWL changes but also speed habituation to them. Finally, the stability or shift upwards in quit-related measures in England over the last three waves suggests some non-HWL activity in England drove responses, especially compared to Australia. These unexpected findings require deeper exploration. Finally, we cannot rule out some of the differences between countries due to differences in response tendencies by country culture, such as understating or choosing stronger terms, but they cannot explain the differences in patterns. Thus, it is premature to conclude that warnings are overall losing potency, although this remains a plausible explanation of our findings.

This study included four high-income countries with strong records of anti-smoking policies and public education around smoking. The findings may not translate into different socio-economic contexts or in populations with little other public education about the harms of smoking. There is some evidence that reactions to HWLs may be stronger in lower-income countries.^19,30 We found evidence of differential rates of wear-out across our four high income countries, which suggests that the rate of wear-out might be due to a range of factors other than country income, extent of other tobacco control activity being a likely candidate.³⁰ Our study did not examine the effects of HWL content and rotation and number of messages. That said, Australia rotates sets of warnings annually and reactions dropped sharply, arguing against consistent benefits. We also cannot draw any conclusions about HWL’s impacts on preventing people from smoking.

Conclusions

HWLs remain an essential tobacco control intervention to raise awareness of tobacco-related harms and discourage use, and they should be much stronger than for any other consumer product given smoking’s extraordinary harms. However, we should temper expectations of their long-term ability to increase cessation, particularly among people smoking daily. Strengthening HWLs can have marked temporary positive effects, but the added benefit, as assessed by the measures used, declines over time, and residual impacts are quite low, more so among people smoking daily. Our findings strongly suggest that reactions to HWL are a function of the change from pre-existing warnings, not their absolute magnitude. Strengthening HWLs appears to only slow, or temporarily reverse, the rate of decline in the desired reactions studied here.

Author Contributions

Bibha Dhungel (Conceptualization [equal], Data curation [lead], Formal analysis [lead], Investigation [equal], Methodology [equal], Project administration [equal], Resources [equal], Validation [equal], Visualization [equal], Writing—original draft [equal], Writing—review & editing [equal]), Ron Borland (Conceptualization [equal], Investigation [equal], Methodology [equal], Project administration [equal], Resources [equal], Validation [equal], Visualization [equal], Writing—original draft [equal], Writing—review & editing [equal]), Hua-Hie Yong (Methodology [equal], Validation [equal], Writing—review & editing [equal]), Coral Gartner (Methodology [equal], Validation [equal], Writing—review & editing [equal]), Katherine East (Methodology [equal], Validation [equal], Writing—review & editing [equal]), David Hammond (Methodology [equal], Validation [equal], Writing—review & editing [equal]), K Michael Cummings (Methodology [equal], Validation [equal], Writing—review & editing [equal]), Andrew Hyland (Methodology [equal], Validation [equal], Writing—review & editing [equal]), Maansi Bansal-Travers (Methodology [equal], Validation [equal], Writing—review & editing [equal]), Ann McNeill (Methodology [equal], Validation [equal], Writing—review & editing [equal]), Shannon Gravely (Methodology [equal], Validation [equal], Writing—review & editing [equal]), and Geoffrey T. Fong (Methodology [equal], Validation [equal], Writing—review & editing [equal])

Funding

The ITC Four Country Survey was supported by grants from the US National Cancer Institute (R01 CA100362, R01 CA090955, P50 CA111236 (Roswell Park Transdisciplinary Tobacco Use Research Center), and P01 CA138389), Robert Wood Johnson Foundation (045734), the Canadian Institutes of Health Research (MOP-57897, MOP-79551, MOP-115016), Australia Commonwealth Department of Health and Aging, the Canadian Tobacco Control Research Initiative (014578), the National Health and Medical Research Council of Australia (265903, 450110, 1005922), and Cancer Research UK (C312/A3726, C312/A6465, C321/A11039, C25586/A19540). The ITC Four Country Smoking and Vaping Survey was supported by grants from the US National Cancer Institute (P01 CA200512), the Canadian Institutes of Health Research (FDN-148477), and the National Health and Medical Research Council of Australia (GTN1106451, GTN1198301). Additional support to GTF is provided by a Senior Investigator Grant from the Ontario Institute for Cancer Research, a Senior Prevention Scientist Award from the Canadian Cancer Society Research Institute, and the Canadian Cancer Society O. Harold Warwick Prize. All other authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

Declaration of Interests

KMC has in the past and continues to serve as a paid witness in litigation filed against cigarette manufacturers. GTF has served as an expert witness or consultant for governments defending their country’s policies or regulations in litigation. DH has served as a paid expert witness on behalf of governments and public health authorities in legal challenges against tobacco and vaping companies.

Data Availability

In each country participating in the International Tobacco Control Policy Evaluation (ITC) Project, the data are jointly owned by the lead researcher(s) in that country and the ITC Project at the University of Waterloo. Data from the ITC Project are available to approved researchers 2 years after the date of issuance of cleaned data sets by the ITC Data Management Centre. Researchers interested in using ITC data are required to apply for approval by submitting an International Tobacco Control Data Repository (ITCDR) request application and subsequently to sign an ITCDR Data Usage Agreement. The criteria for data usage approval and the contents of the Data Usage Agreement are described online (http://www.itcproject.org).

Ethics

The ITC Four Country (Australia, Canada, United Kingdom, and United States) survey protocols and all materials, including the survey questionnaires, were cleared for ethics by Research Ethics Board, University of Waterloo, Canada (REB#10556, REB#13978, REB#17469), Human Research Ethics at Cancer Council Victoria, Australia (IRB HREC 0211), Office of Research Subject Protection, Roswell Comprehensive Cancer Center, US (IRB NT 02-20), and Research Ethics Office, King’s College London, UK (IRB PNM/13/14-151). The ITC Four Country Smoking and Vaping (Australia, Canada, England, and United States) survey protocols and all materials, including the survey questionnaires, were cleared for ethics by Research Ethics Board, University of Waterloo, Canada (REB#20803/30570, REB#20803/30878), Human Research Ethics at Cancer Council Victoria, Australia (IRB HREC 1603), Human Research Ethics Committee, The University of Queensland, Australia (IRB #2016000330), Research Ethics Office, King’s College London, UK (IRB RESCM-17/18-2240), and ethics clearance at the Medical University of South Carolina, US was waived due to minimal risk.