Use of non-invasive wearables biomonitoring to detect presmoking, smoking and postsmoking stages: protocol for an observational laboratory study

Abstract

Introduction

Reliable detection of cigarette smoking is necessary for just-in-time adaptive smoking cessation support. Smoking detection typically relies on intervention recipients to self-report smoking behaviours and their antecedents, which is burdensome and subject to reporting biases, or on specialised sensors and wearables to detect smoking gestures, which may not be feasible for real-world implementation. Here, we describe an observational laboratory-based study protocol designed to identify signature biomarkers and hand–mouth gestures associated with presmoking, smoking and postsmoking using off-the-shelf wearable devices.

Methods and analysis

30 non-treatment seeking individuals who use combustible tobacco products will participate in the study. Participants will be monitored for 1 hour in a smoking chamber, during which time they will wear a Garmin device that will collect hand/arm movement, heart rate, heart rate variability, blood oxygenation and respiratory rate. First, participants will be in a nicotine-deprived state based on 12-hour abstinence validated by exhaled carbon monoxide readings (~20 min). Then, participants will be allowed to smoke cigarettes of their choice (~10–15 min). Finally, participants will be in a nicotine-satiated state (~25 min). Participants will be video recorded to allow us to label the data corresponding to the smoking stage and behaviour. We will conduct time-series analysis and analysis of variance to quantify changes in biomarkers between smoking stages.

Ethics and dissemination

The Institutional Review Board of the University of Kansas Medical Centre approved the study on 21/3/2025 (STUDY00161139). Participants will provide informed consent to participate in the study. Data collection is expected to begin in September 2025 and results will be submitted for publication in 2026.

Trial registration number

NCT07067151.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Use of off-the-shelf wearable devices to identify signature biomarkers and hand–mouth gestures associated with smoking stages.

• Passive sensing of smoking can identify times to deliver momentary smoking cessation support, appeal to populations averse to high-burden treatments and boost intervention efficacy.

• Real-world applications for smoking-detection algorithms in wearables and mobile applications.

• Inclusion criteria can limit the participation of people with high-nicotine dependency.

• Unnatural lab setting may affect participants’ behaviour and associated physiological parameters.

Introduction

Smoking remains a public health issue with disparate prevalence and health outcomes. Smoking prevalence is higher than the national average among certain populations, such as people with lower educational attainment (general education development: 30.7%; 0–12 years, no diploma: 20.1%) and low-income earners (18.3%).¹ Disadvantaged populations are less likely to be the recipients of cessation advice, to use cessation support and aids and to successfully quit smoking.² Mobile interventions can deliver standalone or complementary cessation support at low costs and participation burden.²

Just-in-time adaptive interventions (JITAIs) are mobile interventions that are designed to identify the optimal windows of opportunity or vulnerability for cessation support based on the fast and ever-changing factors within the treatment recipients or their environments.³ This necessitates the collection of users’ psychological or physiological states in real time and in their natural environment.⁴ Most ambulatory assessments in smoking research have relied on self-reports using ecological momentary assessments (EMAs, eg, ⁵,¹⁰), which have several shortcomings. First, self-reporting is dependent on active user input, which introduces participant burden that has been linked to decreased compliance and higher attrition,¹¹ goes against people’s preferences for and expectations of minimal user input^{12 13} and shows reduced efficacy compared with system-driven interventions.¹⁴ Second, self-reporting focuses on psychological factors (eg, motivational states)¹⁵ that are subject to cognitive biases and rest on the assumption that people are capable of introspection.¹⁶,¹⁸ Third, the validity of data obtained through self-reporting can be questionable when the sampling frequency is insufficient to understand the phenomenon under investigation, the time lag between an event and its reporting is too long, or when self-reporting cannot be objectively verified.^{15 19}

Mobile wearable sensors (eg, smartwatches) allow for unobtrusive and continuous capture of data. Beyond the requirement to wear these devices, they have many advantages.⁴ Participant burden, recall bias, social desirability and assessment reactivity are reduced or eliminated. More importantly, wearable sensor data are objective, momentary and ecologically valid. Body-worn sensors have been used to passively detect smoking episodes, lapses or their antecedents either alone or in combination with user input.²⁰ For example, camera-based and thermal-sensor wearables^{21 22} and acoustic sensors²³ can detect smoking events. Motion-based sensors have been used to detect smoking gestures in laboratory and real-world settings.²⁴,³⁰ Other efforts to detect smoking events rely on biomarkers (eg, respiration³¹ or combined smoking gestures with biomarkers).^{32 33} Additionally, there are efforts to detect smoking antecedents. For example, AutoSense relies on a suite of six sensors to detect psychological stress, a risk factor for smoking lapse,^{34 35} and has been used in two JITAIs that delivered stress-management and mindfulness treatments to users attempting to quit.^{36 37} Efforts to detect smoking have relied on specialised, often multisensor, wearables (eg, ^{3132 38},⁴²) that limit their scalability, acceptability and usability outside of lab-controlled settings.²⁰ Fewer attempts relied on commercial wearables to detect smoking using mainly hand/arm movement,^{27 28 43 44} which can leave out other parameters that can improve the detection or prediction of smoking episodes. Finally, limited attempts exist to identify physiological states associated with pre- and postsmoking using wearables-based biomarkers.

Commercially available wearables capture biomarkers that can be useful in detecting smoking episodes, their prior and subsequent states in free-living environments. These biomarkers include heart rate and heart rate variability (HRV), blood oxygenation, respiratory rate, among others. Cigarette smoking is associated with lower HRV⁴⁵,⁴⁸ and higher (resting) heart rate.⁴⁹,⁵³ Smoking is also associated with decreased blood oxygen saturation because carbon monoxide (CO), which studies have shown increases with smoking,⁵³ binds with haemoglobin in the red blood cells and reduces their capacity to carry oxygen.⁵⁴ Finally, smoking involves deep inhalation and exhalation patterns that interfere with breathing, allowing air to flow through the nose only in between puffs with oronasal inhalation and oral exhalation.⁵⁵ Moreover, commercial wearables enjoy increasing acceptability and use among the general population. For example, 35% of US adults use a digital device to track their health,⁵⁶ which makes them suitable candidates for monitoring of smoking in natural settings among wider populations, especially underserved populations.⁵⁷

The objective of this study is to identify wearables-based digital biomarkers that are associated with nicotine deprivation (ie, presmoking), nicotine satiation (ie, postsmoking) and with a smoking episode (ie, during smoking). Identifying signature digital biomarkers of smoking can help us to unobtrusively and accurately detect the proclivity to smoke, which can inform the delivery of momentary support to pre-empt lapsing or progressing towards a relapse among people attempting to quit smoking.^{58 59} The study will help us develop a just-in-time component of a smoking cessation mobile intervention, called Quit Journey, that targets individuals with low socioeconomic status (SES) who smoke cigarettes.

Methods and analysis

Design and hypothesis

This is an observational study of wearable-based markers associated with smoking. Relative to presmoking, we hypothesise that (a) blood oxygen saturation will be lower, (b) heart rate will be higher, (c) HRV will be lower and (d) respiratory rate will be lower during and after a smoking event.

Participants

We will recruit 30 non-treatment seeking participants with low SES who smoke.⁶⁰,⁶² To be included in the study, participants must be 21 years or older, in compliance with Tobacco 21 laws; be maximally high school educated (or equivalent) and earn at or below 200% of the 2024 federal poverty line corresponding to number of people in their household⁶³; have smoked 100 cigarettes in their lifetime and 5+ cigarettes per day in the past year; willing to abstain from smoking for ≥12 hours prior to the lab visit verifiable by a <10 parts per million (ppm) exhaled CO reading obtained during the in-person screening and retrospectively by absence of nicotine in blood sample^{64 65}; willing to abstain from using any nicotine products or devices for ≥12 hours prior to the lab visit; willing to abstain from alcohol use for ≥12 hours prior to the lab session verifiable by breath alcohol concentration=0.000; willing to adhere to all study procedures, including agreeing to a blood draw and bringing ≥1 cigarette for use during the lab visit; have a smartphone (iPhone iOS 14 and higher, Android 9.0 and higher) and willing to download study mobile application onto their phone; speak English and willing and able to provide informed consent. Participants will be excluded if they (a) are not abstinent before the lab session through biochemical verification after two attempts, (b) have past 3-month pulmonary or cardiovascular event, or those who have been hospitalised or visited the emergency room for seizure, stroke or new heart problems, (c) are pregnant or breastfeeding, (d) report quit attempts in the last 2 months or intend to quit in the next month and (e) are unable or unwilling to provide informed consent. We will strive to have representation across racial and/or ethnic groups and biological sex.

Recruitment

Potential research volunteers (PRVs) will be recruited using standard methods, including flyers, social media and from past study participant pools. Additionally, we will use low-touch recruitment methods, including querying the electronic health record for health system patients that have opted in to be contacted for research studies and whose records indicate current smoking. PRVs will be sent a letter informing them about the study and that a research staff member will reach out to further discuss the study. PRVs will be contacted by research staff via phone and/or text message within 7 days of receiving the letter and offered screening.

Study procedures and informed consent process

Interested PRVs will contact the study team by phone, text message or email, depending on their preference. The study team will screen PRVs for eligibility to enrol in the study by phone or via online self-screening. This screening is preconsent. Eligible participants will have an in-person screening to establish their smoking status (ie, exhaled CO is ≥10 ppm) and provide written informed consent. The study team will review the informed consent, which states that this is a research study, outlines all study activities and duration and emphasises that participation is voluntary and confidential. Participants will be informed of potential risks associated with the study, including physical risk from blood draw, psychological risk in response to the study questions and the potential for data breach.

Individuals who provide informed consent to participate in the study will be assigned a participant ID and asked to complete a baseline survey. At their in-person screening, they will schedule their lab visit and receive a Garmin device to wear for 3 days prior to their lab visit. The study team will help participants instal the Catalyst app that will allow us to capture the wearables data. On the day of their lab visit, we will biochemically verify that participants are abstinent via a Micro+Smokerlyzer (Bedfont) CO monitor.⁶⁶ Participants must have a reading <10 ppm. Additionally, we will obtain a blood sample and exclude from the analysis participants who fail abstinence verification based on blood nicotine level (<3 ng/mL).

Those who are verified as abstinent will proceed to the lab session where they will be escorted to a smoking room while wearing two Garmin devices, one on each arm, to avoid data loss. Throughout the lab visit, participants will also wear a cuff for blood pressure and heart rate monitoring, a pulse oximeter for monitoring blood oxygen saturation, pulse rate and respiratory rate. We will monitor each participant for 1 hour during which they will remain in a nicotine-deprived state (~20 min presmoking), then will be asked to smoke cigarettes of their brand of choice (~10–15 min smoking), then continue in a nicotine-satiated state (~25 min postsmoking). Participants will be seated during the lab session. The 1-hour lab session will be video recorded to allow us to label the data for each of the three stages of the study and based on participants’ behaviour during the smoking episode. Participants will answer a brief pre- and postsession survey. The schedule of activities for the study appears in table 1.

Table 1. Schedule of activities.

	Pre-enrolment	Enrolment, day 0	Home monitoring	Lab session
				Presession	Nicotine deprived (~20 mins)	Smoking episode (~10–15 mins)	Nicotine satiated (~25 mins)	Postsession
Initial screening	x
In-person screening		x
Informed consent		x
Baseline survey		x
Eligibility verification (CO, serum)		x		x
Pre- and post-lab session survey				x				x
Wearables data			x		x	x	x
Medical vitals				x	x	x	x	x
Video recording					x	x	x

CO, carbon monoxide.

The lab session will take place at the University of Kansas Medical Centre (KUMC) that has dedicated smoking rooms under negative pressure. The rooms are in the Clinical Research Unit (CRU) and are equipped with medical-grade instruments to measure heart rate, blood pressure and blood O₂ by pulse oximetry. Furthermore, the CRU is staffed by research nurses and is fully equipped for phlebotomy procedures and will allow for blood collection before the lab session. Participants will receive an e-gift card for up to $150.

Study evaluations

First, the phone/online screening questionnaire will include inclusion- and exclusion-related questions (table 2). Prior to beginning the lab session, participants will undergo biochemical abstinence verification procedures measured by exhaled CO readings. The baseline blood draw will be used to verify abstinence post hoc. Participants shown to have used nicotine prior to the visit will be removed from final analysis. Self-report will be used for non-combustible nicotine product abstinence, as there is currently no objective and reliable measure to verify non-use of these products within the time participants will be present in the lab. Second, the baseline survey will include questions on demographics, social determinants of health, smoking history and nicotine dependence,⁶⁷ other tobacco and substance use, access to healthcare and skin colour/tone.⁶⁸ During the 3 days of device wearing, participants will record every smoking episode, activity and mood on the Catalyst app.^{10 69 70} During the lab session, we will use Garmin devices to collect heart rate, HRV, respiratory rate, blood oxygenation levels and accelerometer and gyroscope data that will be used to capture hand-to-mouth gestures. We will also capture medical-grade data (ie, blood pressure, heart rate, blood oxygen saturation, pulse rate and respiratory rate). Finally, prior to and following the lab session, participants will answer a brief questionnaire on withdrawal,^{71 72} smoking urges,^{7 73} negative and positive affect,⁷⁴ mood, cravings and smoking expectancies.^{6 70 75}

Table 2. Study measures.

Domain	Measure	Sampling frequency
Demographics and social determinants of health	Age, sex, race and/or ethnicity, education, family size, annual income, marital status, employment, job security, financial security, language proficiency and language spoken at home, health literacy and numeracy	1
Smoking behaviour and history	Smoking status, number of daily cigarettes smoked, smoking initiation, cigarette info (eg, type, cost, brand and flavour), nicotine dependence, quit attempts and other tobacco use	1
Health factors	Disability, underlying medical conditions, health insurance, primary care clinician, general health, weight, waist circumference and substance use	1
Smoking episodes	Recording of smoking episodes and associated state and context during 3-day home monitoring	Participant dependent
Wearables data
Heart rate	Continuous heart rate measure	continuous
Resting heart rate	Average heart rate when at rest	continuous
HRV	Variation in time interval between heartbeats	continuous
Respiratory rate	Number of breaths per minute	continuous
Blood oxygenation	SpO2 levels indicating oxygen saturation in the blood	continuous
Accelerometer data	Acceleration along X, Y and Z axes (3-axis acceleration)	continuous
Gyroscope data	Rotational movement along X, Y and Z axes (3-axis angular velocity) and degree of rotational movement (rotation) estimated by integrating angular velocity over time	continuous
Medical vitals
CO	Micro+Smokerlyzer (Bedfont)	3
Serum	Nicotine, cotinine, hydroxycotinine, measured by gas chromatography-mass spectrometry	1
Heart rate	iProven blood oxygen monitor fingertip	continuous
Blood oxygenation	iProven blood oxygen monitor fingertip	continuous
Blood pressure	Spot vital signs 4400 (Welch Allen)	continuous
Respiratory rate	iProven blood oxygen monitor fingertip	continuous

CO, carbon monoxide; HRV, heart rate variability.

Analyses

Video recordings will be labelled to mark the start and end of the nicotine-deprived state, the smoking episode and the nicotine-satiated state. During the smoking bout and for each puff, we will label its start and end, which will give us information on the number of puffs, duration of puff and time between puffs.

We will employ several statistical methods to quantify changes in biomarkers independently and collectively across the presmoking, smoking and postsmoking stages. Initially, descriptive statistics, including means, SD and ranges, will be calculated for each biomarker in each stage. Effect size measures, such as Cohen’s d, will be calculated for changes in biomarkers between stages. Time-series plots for each biomarker will be created to visualise trends across the three stages. All data will be checked for normality. Assuming the homogeneity of variance, we will perform repeated measures analysis of variance separately for each biomarker to assess significant differences across the three stages. If significant differences are found, post hoc pairwise comparisons will be performed to identify which stages differ significantly. To account for within-subject correlations and potential covariates (eg, age, sex and nicotine dependence), linear mixed-effects models will be developed for each biomarker, with stage as a fixed effect and participant as a random effect.

Change point detection algorithms, such as the Pruned Exact Linear Time (PELT) algorithm,⁷⁶ will be applied to identify significant shifts in biomarker levels, potentially indicating transitions between stages. To assess the collective changes in all biomarkers across stages, multivariate analysis of variance will be performed. The 1-hour period of data collection will be divided into 10 s epochs. Each epoch will be labelled as one of the three possible stages (ie, presmoking, smoking and postsmoking) based on the video recordings. For each epoch, in addition to the physiological biomarkers, we will incorporate features derived from the accelerometer and gyroscope data to capture smoking-related hand movements. These movement features will include time-domain features (mean, SD, root mean square, peak-to-peak amplitude and zero-crossing rate), frequency-domain features (dominant frequency and spectral energy derived from Fast Fourier Transform) and orientation features (pitch, roll and yaw angles) for each axis of both accelerometer and gyroscope data. These movement features will represent the characteristic hand-to-mouth gestures associated with smoking, providing additional behavioural context to complement the physiological data.

The combined dataset of physiological biomarkers and movement features will serve as input for training and validating machine learning models. We will implement machine learning approaches, including random forest and support vector machine classifiers, to predict smoking stages based on this multimodal data. The dataset will be partitioned into training (80%) and testing (20%) sets. Cross-validation techniques, such as k-fold cross validation, will be employed to ensure robust model performance and to mitigate overfitting. All statistical analyses will be performed using R and Python, with a significance level set at 0.05. Multiple comparison corrections will be applied where appropriate to control for familywise error rates.

Data security

We will have a smartphone dedicated to the study’s lab visit on which we will instal the Metricwire Catalyst app. Participants will wear two Garmin devices during the lab session that will allow us to capture raw sensor data, including the gyroscope data. The Metricwire Catalyst app interacts with the Garmin Standard Software Development Kit (SDK), allowing the app to pair directly with the Garmin device via Bluetooth and record raw sensor data. To capture wearables data prior to the lab visit, we will assign anonymous email accounts to set up user profiles on the study app and pair them with the Garmin device given to each participant. Data collected are securely synced with the Metricwire Research Portal to ensure that the data are not stored on the smartphone longer than is necessary. Neither personally identifiable information nor protected health information will be known to Metricwire or Garmin. We will download the data at the end of the study to the National Institute on Minority Health and Health Disparities (NIMHD) and KUMC servers.

Patient and public involvement

None.

Ethics and dissemination

The protocol has been reviewed and approved by the scientific review committee of NIMHD. The Institutional Review Board (IRB) of KUMC approved the study on 21/3/2025 (IRB ID: STUDY00161139) and two amendments on 30/5/2025 and 16/6/2025. The IRB of the National Institutes of Health (NIH) reviewed the study to ensure compliance with NIH requirements (IRB002323). The study is registered on Clinicaltrials.gov (NCT07067151). We anticipate data collection to begin in September 2025 and continue for 3 months. Results will be submitted for publication in 2026.

Discussion

Once implemented, this study will generate data on physiological markers associated with smoking stages derived from commercial wearables. Real-time passive identification of smoking events and their preceding and subsequent states will expand our understanding of these phenomena beyond self-reported psychological factors, build on prior work using wearables-based data to detect smoking and can maximise the efficacy of smoking cessation interventions.⁷⁷ Specifically, the detailed characterisation of presmoking, smoking and postsmoking states could inform intervention strategies for each phase.³ The study will inform the development of just-in-time momentary support to pre-empt a lapse or prevent it from progressing to a relapse for the users of Quit Journey, a smoking cessation intervention targeting individuals with low SES. This study will also increase the representation of underserved populations, help more accurately develop algorithms for smoking detection that include input from minority populations and inform a subsequent real-world study to detect and characterise smoking behaviour in natural settings.⁷⁸

Disadvantaged populations prefer self-help technology-based cessation interventions, such as mobile apps.⁷⁹,⁸¹ Evidence among racial and/or ethnic minorities shows near saturation of mobile phone ownership, growing trends in ownership of commercial smartwatches and fitness devices and willingness to share wearables data for health purposes.⁸²,⁸⁵ Collectively, these data support the development of effective mobile-based cessation interventions to aid underserved populations in their quit attempts and reduce smoking-based health disparities. However, smoking cessation interventions that target underserved populations, including technology-based interventions, are few, have modest or inconsistent effects and their methodological rigour can be improved.⁸⁶,⁸⁸ This protocol is a part of a systematic effort to develop and evaluate a smoking cessation intervention, Quit Journey, following the multistage optimisation strategy.⁸⁹

Passive sensing of smoking events by mobile wearable sensors can improve the efficacy of personalised interventions for several reasons. First, passive sensing can provide an unobtrusive and objective account of smoking events and their antecedents that can be used to deliver timely cessation support.⁴ Studies show that tailored interventions are more effective than generic or targeted interventions in inducing behaviour change⁹⁰ and that individual-level predictors are better at predicting lapse and smoking behaviour than group averages.⁹¹ However, there are few JITAIs for smoking cessation,⁹² mainly providing location-based⁹³ or stress-timed support³⁷ informed by self-reports prior to or during high-risk smoking situations.⁹⁴,⁹⁶ Limited efforts to provide wearables-based timely notifications or support to users exist.^{25 27 97} Despite the barriers that disadvantaged populations encounter when attempting to quit smoking,⁹⁸ we are aware of one intervention that targets individuals of low SES, the Smart-T app, which relies on EMAs to collect data on lapse risk (eg, urge to smoking and cigarette availability) to deliver tailored support messages.^{5 99 100} Disadvantaged populations are more likely to experience smoking triggers due to, for example, financial stressors and an abundance of smoking cues in their environment.¹⁰¹ These changing psychological states and environmental exposures are optimal targets for JITAIs targeting disadvantaged populations as a priority group for tobacco control efforts.

This study builds on prior work on smoking detection using specialised non-commercial sensors and wearables.³⁴ Beyond smoking cessation, our efforts parallel a growing trend to use mobile wearable sensors to identify behaviours (eg, eating detection, alcohol drinking and cocaine use)¹⁰²,¹⁰⁴ and their precursor states (eg, stress)¹⁰⁵ in real-life settings. More generally, the machine learning models integrating physiological biomarkers and movement data could be implemented in commercially available wearable devices and smartphone applications. This implementation will make mobile cessation interventions smarter, contrary to the standing critique of cessation apps that employ mostly unidirectional or bidirectional intervention components (eg, psychoeducation).^{106 107} Finally, the data collected from this laboratory-based study will serve as baseline parameters for detecting smoking events in free-living conditions. In a planned study, we aim to fine-tune our smoking-detection algorithms with data collected in natural environments, potentially improving their robustness and generalisability.

Study strengths include its focus on individuals with low SES who have higher smoking prevalence and related health outcomes; reliance on multiple physiological parameters to improve the accuracy of detecting presmoking, smoking and postsmoking stages and use of commercial wearables that are acceptable to users for data collection in natural settings. The controlled lab environment eliminates noise from outside factors that may affect physiological parameters. However, participants might alter their natural smoking behaviour, potentially affecting the physiological parameters we aim to measure.¹⁰⁸ The inclusion criteria may introduce bias by excluding those with high-nicotine dependence who are unwilling or unable to abstain from smoking for 12 hours. This 12-hour abstinence period creates a presmoking condition that may not accurately reflect real-world smoking behaviours where the duration between smoking episodes is often shorter. The 1-hour observation period, while intensive, provides only a snapshot of smoking behaviour and may not capture the full range of physiological variations throughout a typical day. The postsmoking observation time may not capture physiological changes that take longer to occur. We are not examining other behaviours that share commonalities with smoking in hand movement. We are only investigating markers associated with smoking events in a seated position and with smoking a whole cigarette, rather than one or two puffs. Participants’ mood and use of other substances can affect puff duration and interpuff intervals. Participants will smoke their choice of cigarette brand, but we will collect comprehensive data on the brand (eg, nicotine and tar content). Although questions arise about the accuracy of wearables-based data, especially when collected from individuals with darker complexions,¹⁰⁹ changes in betas should be informative of changes in physiological markers across the three smoking stages.

In conclusion, this protocol outlines the aim and study procedures of a planned lab-based study that will identify commercial wearables-based biomarkers associated with smoking events and their precursor and successor states. Passive detection of smoking episodes using wearable technologies can potentially boost the efficacy of mobile smoking cessation interventions among populations disproportionally affected by smoking and its health outcomes and facilitate the implementation of smoking interventions in naturalistic settings.

This research was supported [in part] by the Intramural Research Program of the National Institutes of Health (NIH). The contributions of the NIH author(s) were made as part of their official duties as NIH federal employees, are in compliance with agency policy requirements, and are considered Works of the United States Government. However, the findings and conclusions presented in this paper are those of the author(s) and do not necessarily reflect the views of the NIH or the U.S. Department of Health and Human Services.