Day-to-Day Variability in Self-Reported Cigarettes Per Day

Abstract

Background and Aims

Nicotine addiction theory predicts small day-to-day variability in cigarettes/day (CPD) whereas social learning theory predicts large variability. A description of the variability in CPD over multiple days is not available.

Methods

We conducted secondary analyses of two natural history studies with daily smokers—one of smokers not intending to quit, and one of smokers intending to quit sometime in the next 3 months. In the former, smokers recorded their smoking during the day by Ecological Momentary Assessment, using a palm-top computer. In the latter, participants reported CPD nightly via a phone Interactive Voice Response system. Analyses were based on smokers who reported averaging ≥10 CPD, and on days in which there was no attempt to stop or reduce smoking.

Results

Across the two studies, on average, smokers had small changes in day-to-day CPD (mean changes were 2.2 and 2.9 CPD). However a minority averaged changing by ≥5 CPD from one day to the next (7% and 11%), and many changed by ≥5 CPD on at least 10 of the 90 days (8% and 31%). Neither smoking restrictions, dependence, stereotypy ratings, nor interest in quitting predicted variability.

Conclusion

Although on average, smokers have little change day-to-day CPD, a substantial minority of smokers often change by 5 CPD from day-to-day. We did not find potential causes of this variability.

Implications

Across day variability in CPD is larger than implied in prior studies. Determining causes of day-to-day variability should increase our understanding of the determinants of smoking.

Introduction

Several studies have reported on the variability in cigarettes/day (CPD) over time within a day.¹ Others have reported CPD variability across months and years.² Unexpectedly, we could locate only two prospective reports of variability of CPD across consecutive days.^3,4 Given this paucity of evidence, we undertook a secondary analysis of two datasets from non-treatment cohort studies to examine the amount of day-to-day variability in CPD, and predictors of CPD variability.

Determining the day-to-day variability in CPD is important for several reasons. First, CPD is often used both as an independent variable to measure exposure to toxin intake, nicotine dependence, etc., or as a dependent variable to measure success of tobacco policies, treatment programs, etc. Many studies ask smokers CPD over periods of the last week, month, etc. These retrospective reports suggest little day-to-day variability in smoking; for example, in one study half of smokers gave the exact same CPD for each day of the last week, and most of these reports (81%) were exactly the same as their reported “usual” CPD.⁵ This would be consistent with theories of nicotine dependence. The theories posit much of smoking is driven by smoking to avoid or relieve withdrawal symptoms.⁶ Given this, smokers should smoke at regular intervals to maintain nicotine levels and have little day-to-day variability in CPD. In fact, one manifestation of drug dependence is a “narrowing of the repertoire;” that is, stereotypy or using the same amount of drug independent of the situation.⁷ In contrast, substantial day-to-day variability occurred in the two prior prospective studies.^3,4 This result would be consistent with Social Learning Theory that posits interoceptive and environmental cues (which vary from day-to-day) control smoking behavior.⁸

One of the two prior studies examined changes in CPD over 21 days using Time-Line-Follow-Back (TLFB),³ however, retrospective recall over such long periods is known to underestimate variability.⁵ The other study examined day-to-day CPD change using daily reports among smokers applying for non-treatment laboratory studies but only over 3 days.⁴ We conducted a secondary data analysis from two datasets of large prospective, observational cohort studies that examined CPD daily, over multiple days to provide a more detailed description of day-to-day variability in CPD and the variables associated with increased variability.

Methods

The Ecological Momentary Assessment (EMA) Dataset

The prior study described above that reported CPD using TLFB also measured CPD using EMA. This study included daily and non-daily smokers who were not trying to quit. Our analysis focused on the EMA data from daily smokers. The study had smokers record each cigarette for 21 days, with provision for recording “missed” entries at end of day. No treatment was provided. The study is described in more detail elsewhere.³

The Interactive Voice Response (IVR) Dataset

The other dataset came from a pilot and main study that used daily IVR monitoring to describe reduction and quit attempts among daily smokers. To obtain smokers likely to quit or reduce during the study period, the study included smokers who intended to quit smoking at some point in the next 3 months. Each night, participants called the IVR system to report CPD, whether they were trying to quit or reduce, and other outcomes. The methods of the pilot and main study were identical except the pilot lasted 1 month and the main study 3 months. Exploratory data analysis suggested the results of the two studies were very similar; thus, we pooled the data from these studies for analysis. The two studies are described in detail elsewhere.^9,10

Data Analysis

The EMA study did not include smokers who intended to quit in the near future. The IVR sample did include such smokers; thus, in that dataset, we used only days of smoking in which the smoker did not plan to or had not tried to stop or reduce smoking. We also omitted the 7 days following a quit attempt—to allow CPD to stabilize after a quit attempt.¹¹ In addition, to have a sufficient sample, in both the EMA and IVR data sets we limited analysis to participants who had ≥10 pairs of consecutive days. The IVR sample required participants smoke ≥10 CPD but the EMA study did not. To make the two samples more similar, we only included those in the EMA sample whose “usual” CPD during the study was ≥10. In the EMA study, 38 (20%) of participants did not contribute sufficient data for analysis or had baseline CPD < 10; in the IVR study, 70 (36%) were excluded for insufficient data. In the EMA study 6% of days were missing, leaving 3081 days. In the IVR study, 5% of the days were missing and we excluded 31% of days due to quit/reduction attempts, leaving 6022 days.

The participants used in the analysis were mostly middle-aged, white smokers who smoked about a pack of cigarettes a day, and were of moderate nicotine dependence (Table 1). The two samples differed from the average US smoker¹² and from each other in several respects.

Table 1.

Study and Participant Characteristics

	Smokers trying to quit	Smokers not trying to quit	Average US daily smoker, 2007a
Sample size	122	156
Age	44 (12)	42 (11)	44a
% women	67%	43%***	46%a
% minorities	20%	35%**	22%a
% employed	53%	64%*
Mean (SD) usual CPD	19 (10)	17 (6)	16a
FTCDb	5.3 (2.4)	5.4 (1.7)	4.5b
Days of monitoring	28/84c	21
Monitoring method	Nightly IVR	EMA

CPD = cigarettes/day; EMA = Ecological Momentary Assessment; FTCD = Fagerstrom Test for Cigarette Dependence; IVR = Interactive Voice Response; SD = standard deviation.

^aHughes & Callas, 2010²⁵

^bFagerstrom & Furberg, 2008²⁶

^cPilot/main study.

Smokers trying vs. not trying to quit

***p < .0001.

**p = .004.

*p = .05.

We examined two measures of variability. The first was mean successive difference (MSD) which was the mean of the absolute changes in CPD from one day to the next within a participant.¹³ The second was the coefficient of variation (CV) which is the within-subject standard deviation divided by the within-subject mean. The MSD measures absolute change in CPD, whereas the CV measures variability in CPD in proportion to the mean CPD. For descriptive purposes we report the 25th percentile, median, and 75th percentiles of both measures.

Results

Overall CPD decreased slightly across days in the EMA study (median decrease of 1.4 cigarettes/week), and even less so among smokers in the IVR sample (0.05 cigarettes/week). The inter-quartile range (middle 50%) for MSD was 1.6 for the EMA and 1.8 for the IVR samples. The inter-quartile range for CV was 0.16 for the EMA and 0.07 for the IVR samples.

Across both samples, the average smoker changed his/her CPD by an average of about 2–3 CPD from one day to the next (Table 2). In the EMA sample, 11% of participants averaged changing by ≥5 CPD from one day to the next and in the IVR sample 7% did so. In addition, 8% of the EMA sample changed CPD by ≥5 CPD on ≥10 of the 90 days of the study, and 31% of the IVR sample also did so. To better illustrate whether the above results represent a large or small amount of variability, we graphed the results of the smoker whose MSD was closest to the middle of the lowest quartile, the smoker whose MSD was closest to the median, and the smoker whose MSD was closest to the middle of the highest quartile for each sample (Figure 1). The Figure 1 shows that even participants with low variability still vary their consumption from day-to-day, while participants with high MSD show very frequent and sometimes dramatic changes in cigarette consumption.

Table 2.

Average Variability by Measure and by Sample

	25th percentile	Median	75th percentile
EMA study
MSD	2.3	2.9	3.9
CV	0.22	0.30	0.38
IVR study
MSD	1.6	2.2	3.4
CV	0.13	0.17	0.24

CV = coefficient of variation; EMA = Ecological Momentary Assessment; IVR = Interactive Voice Response; MSD = median successive difference.

Figure 1.

Cigarettes/day among smokers closest to the middle of the lowest quartile, the median and the middle of the highest quartile for the median successive difference.

We next tested whether baseline characteristics predicted MSD or CV via univariate regressions. In the EMA sample, the MSD was slightly but significantly higher in smokers with higher baseline CPD (F = 27.1, p < .001) and a similar, non-significant trend occurred for the IVR sample (F = 3.5, NS). In both samples, the CV was less in those with a higher baseline CPD in the <20 CPD range, but did not continue to decline in those whose CPD was ≥20 (F = 9.9 in the EMA sample and F = 9.4 in the IVR sample). Neither age, sex, race/ethnicity (white/non-Hispanic vs. other), employment (full or part time vs. other). Fagerstrom Test of Cigarette Dependence¹⁴ (without CPD question), smoking restrictions, self-rated addiction, nor Mental Health Inventory¹⁵ score consistently predicted variability across measures and samples. In the EMA sample, neither total NDSS¹⁶ nor NDSS stereotypy subscale score predicted MSD or CV. In the IVR sample, interest in quitting did not predict variability.

Discussion

Our main finding is that, on average, smokers typically change their cigarette consumption by 2–3 CPD from one day to the next, and that a substantial minority of smokers change by 5 CPD or more on many days. Neither smoking restrictions, dependence, stereotypy ratings, nor interest in quitting predicted variability.

One liability of our analyses is the samples we used for analyses are not representative of all US smokers (see Table 1). For example, we did not include non-daily smokers or those who smoked < 10 CPD daily, even though these represent 22% and 17% of smokers.¹⁷ Second, smokers were asked to record CPD daily in the IVR sample and multiple times daily in the EMA sample. The real-time monitoring in the EMA study is likely why it showed more variability than the daily recall in the IVR study. Third, monitoring of CPD could be associated with “reactivity”, suppressing cigarette consumption; however, the three randomized, controlled trials of daily monitoring among smokers not actively trying to quit found little evidence of reactivity,^18–20 perhaps because reactivity appears to be less when participants are not actively engaged in a change attempt, and when little experimenter contact occurs.^21,22 Fourth, we had no biological verification of changes in day-to-day CPD because this would have required daily biological sampling. In any case, most biomarkers would be insensitive to these changes²³; thus, it is unclear how much of the variation in CPD was due to actual variation in CPD versus variation in reporting CPD. For example, the finding that the variability was smaller in the IVR sample than in the EMA sample could be due to failure of the EMA participants to record some cigarettes smoked. Notably, participants in the EMA study had an end-of-day opportunity to recall those missed cigarettes, much as though they were doing IVR reporting. One asset of the study was the large number of participants and the large number of observations for each participant. Another asset was the use of two different samples to assess convergent validity. Finally, the studies had few drop-outs and small amounts of missing data.

Our study replicated the main findings of the two prior studies. The study that examined daily reports of CPD over 3 days⁴ found that many (40%) smokers changed by ≥5 CPD from one day to the next at some point during the 3 days. The study also found that greater dependence predicted less variability, which we did not find. The other study reported that examined TLFB data reported an across-day standard deviation of 3.0 CPD.³

Our results have practical and theoretical implications. The practical implication is that the amount of variability in the current study suggests some smokers would find it difficult to provide a single number to describe their average CPD. Given this, one would expect that using IVR or EMA may be needed to provide a more reliable estimate. This would be especially important if reduction in CPD rather than abstinence were the outcome of interest; however, over what time period this would need to be collected is unclear. Our two studies found very little change over time across weeks, suggesting collecting this data even over a single week could be valuable.

In terms of theory, classical conceptualizations of drug dependence posit that most cigarette use is to achieve a drug level that would avoid withdrawal.⁶ This conceptualization leads to the expectation that drug use would occur at very regular intervals. Our results suggest that although this conceptualization may apply to some smokers (eg, those in the lowest quartile in the Figure 1), many smokers have substantial day-to-day variability in CPD. This variability is consistent with the large amount of data that indicates environmental and internal stimuli can control smoking behavior.²⁴ The question remains as to what degree these stimuli can account for day-to-day variability. For example, we thought that smokers who had higher perceived tobacco control, more smoking restrictions, and more interest in quitting, would have greater variability, and those that were more dependent and had greater stereotypy scores would have lower variability; but we did not find this. However, there are several other potential influences which we did not analyze; for example, how many hours awake and how much alcohol consumed.

In summary, many smokers showed substantial day-to-day changes in CPD. Identifying causes of this variation could suggest interventions to reduce CPD. Field studies of longer duration could especially be useful. One possible method to explore day-to-day variability would be to monitor CPD daily, identify times when CPD changed dramatically (eg, by >5 CPD) and ask smokers why this change occurred.

Funding

This study was funded by US National Institutes of Health grant DA-025089.

Declaration of Interests

JRH has received consulting and speaking fees from several companies that develop or market pharmacological and behavioral treatments for smoking cessation or harm reduction, and from several non-profit organizations that promote tobacco control. He also consults (without payment) to Swedish Match. None of the other authors have any relevant disclosures.