Worldwide, tobacco use is a major risk factor for disease and death. Tobacco dependence, which is classified in the International Classification of Diseases (ICD-10) [1
], causes several types of pulmonary and cardiovascular illness (emphysema, chronic bronchitis, heart attacks and strokes), lethal cancers (lung, colorectal, mouth, larynx, liver, cervix, etc.), and is known to affect the reproductive and immune systems [2
]. It also increases the chance of severe health issues like diabetes, duodenal ulcers, loss of appetite, atherosclerosis, age-related macular degeneration and vision loss, premature birth and even miscarriages in pregnant women [5
]. Cigarette smoking is the predominant form of tobacco use.
In 2017, an estimated 47.4 million U.S. adults (19.3%) were reported as using tobacco products, including cigarettes (14.0%; 34.3 million); cigars, cigarillos, or filtered little cigars (3.8%; 9.3 million); electronic cigarettes (e-cigarettes) (2.8%; 6.9 million); smokeless tobacco (2.1%; 5.1 million); and pipes, water pipes, or hookahs (1.0%; 2.6 million) [6
]. Research shows that the lifespan of cigarette smokers is generally reduced by 13–14 years [7
]. Also, the toxicants in second-hand smoke (smoke inhaled by people in the surroundings of tobacco smokers), such as carbon monoxide (CO), tobacco-specific nitrosamines (TSNA), formaldehyde (CH2
O), and hydrogen cyanide (HCN), have a deadly impact upon chronic obstructive pulmonary diseases (COPD) and asthma [8
]. According to the United States Centers for Disease Control and Prevention (CDC, Atlanta, GA, USA) estimation, every year about 480,000 deaths are related to first-hand smoking (direct smoke inhalations) and 41,000 to second-hand smoke [9
]. The World Health Organization (WHO) has estimated that annual deaths related to smoking will be 10% (more than 8 million people per year) by 2030 worldwide [10
]. Of those deaths, 75% will be in low- and middle-income countries. There are also substantial economic consequences of smoking. In the United States alone, an annual cost of more than $
300 billion, including $
170 billion for direct medical care and $
156 billion in lost productivity, is generated by 34.3 million adult smokers [11
Despite these statistics, “smoking cessation is often hindered by the low perceivability of health risks and the unawareness of habits in day-to-day life” [12
]. Data from the National Health Interview Survey (NHIS [13
]) suggest that 68% of smokers are interested in quitting, and 85% have attempted quitting at least once in their lifetime [14
], averaging 4 quitting attempts [16
], with 70% of these quit efforts failing eventually [17
]. Although there are numerous treatments available to help people quit smoking [18
], the overall success rates of smoking cessation interventions are low. A critical starting point for these smoking cessation methods is the collection of information on the smoking habits of the individual. Self-reports of the ‘number of cigarettes smoked’ were among the first accepted measures of this information [25
]. These approaches include self-report history methods such as 24-h/7-day retrospective smoking recall [26
], immediate logging of cigarettes after consumption [27
], and instrumented methods, such as ecological momentary assessment (EMA [28
]). Self-report methods have improved in convenience and duration with the increased use of smartphones [29
]. Clinical interventions (nicotine patches [30
], personal counseling [31
], etc.) mostly depend upon these self-report methods to understand smoking habits and estimate the degree of smoke exposure. However, these methods cannot capture detailed smoking metrics (the depth of inhalations, duration of smoke holding, the number of puffs-smoke intake per cigarette, the duration, or other aspects of smoke exposure [35
]), which can support effective interventions and lapse monitoring. Also, as self-report approaches rely heavily on the user’s recall and impose a burden on the smokers [25
], the accuracy of these self-reports is generally limited by memory biases and intentional or unintentional misrepresentations or underreporting [36
During the past decade, a wide range of technology-driven smoking assessments has been investigated, such as expired CO monitoring [37
], biomarkers [39
] and image processing [40
]. However, no usable pattern of inhalations or smoking habits can be drawn from expired CO- or biomarker-based approaches [38
]. A commercial handheld monitoring device, the Clinical Research Support System (CReSS) [43
], was developed to acquire and store behavioral information about smoking in the natural environment. However, the use of this ‘smoke-through’ CReSS device may affect the pattern of inhalations in many smokers due to its obtrusiveness and large size [44
]. Moreover, the ability of this device to capture all instances of smoking does demand that the people being monitored smoke all their cigarettes through the device—not all smokers are compliant with these instructions. Surveillance camera-based imaging methods require the installation of video cameras in all possible smoking locations, which is not feasible at the community level [45
Recently, wearable sensors [46
] have drawn attention as a potential solution to the problem of the passive detection of cigarette smoking and smoke exposure. Wearable sensors are lightweight, mobile, convenient, with the ability for ‘collecting data anytime, anywhere and often’ [47
]. These devices are composed of varying sensing modalities, such as electrical, inertial (individual or multi-axis combinations of precision gyroscopes, accelerometers, magnetometers), acoustic, etc. Some approaches have used a combination of sensors. However, no single wearable method has been found to be 100% accurate for detecting smoking events in all circumstances, isolating puffs and smoke inhalations, or evaluating the metrics of smoke exposure. Some technologies suit certain environments, while others fail to provide good results in the same context. The sensor responses are often influenced by ambient factors, such as motion and clothing. To date, no in-depth survey or comparison (trade-off) study of these approaches has yet been performed highlighting the advantages and limitations of sensing technologies or their applicability in naturalistic settings. Also, there has been little evaluation of the underlying detection algorithms and their comparative accuracy.
This review is intended to provide a systematic evaluation of state-of-the-art wearable sensors for monitoring cigarette smoking in free-living conditions. The primary focus of this review is an up-to-date summary of recent novel approaches, individual and multi-sensor combinations, body locations, processing of sensor signals, detection algorithms and assessments of comfort. To cover the full range of the monitoring systems of cigarette smoking in this survey, research publications and commercially available sensor systems were thoroughly studied, and a total of 314 papers (without duplication) were found related to these topics. Following the application of inclusion and exclusion criteria, 108 papers were selected for a full-text review.
The paper is organized as follows. First, the methodology of the systematic review is presented in Section 2
, along with the specification of the research questions (RQ). Section 3
and Section 4
present the detailed exploration of these research questions, with the identification of the behavioral and physiological manifestations of cigarette smoking (Section 3
), and the evaluation of wearable sensing technologies (Section 4
). Section 5
discusses the challenges and potential research focus in the field of automated monitoring. Section 6
provides a summary of the review.
3. Behavioral and Physiological Manifestations of Cigarette Smoking
The philosophy behind the implementation of wearable sensors in smoking detection requires a thorough comprehension of the cigarette smoking process. The frequency or pattern of cigarette smoking generally varies between individuals or brands of the smoked cigarette; however, a few similarities between behavioral and physiological phenomena are always present [49
]. An average smoker smokes a cigarette in 4–8 min [51
] with 8–16 puffs [52
]. The process starts with the removal of a cigarette from a packet, generally using fingers (sometimes using the combination of teeth and lips), putting the filtered end in the mouth, and lighting up. The number of consumed cigarettes may be tracked from the cigarette packet or holder, if it is instrumented accordingly. Also, smokers usually carry a personal lighter or match, and use it to light their cigarettes. The frequency of cigarette consumption can be identified from cigarette lighting events [12
Once the cigarette is lit, smokers inhale and move their hands away from the mouth. This step is repeated throughout the smoking session. During puffs, the smoking hand stays vertically close to the mouth. Specifically, for inhalations, the fingers holding the cigarette reach closer to the lips and the wrist moves close to the chest. The positioning of these body parts can be used as a potential indicator of smoking events [53
When people pull their hands closer to their mouths (from the rest) for puffing, they need to work against the pull of the acceleration due to gravity. When the hand remains stationary, close to the lips, this gravitational acceleration stays constant. When the hand returns after puffing, it works along with gravity. Smoking puffs can be identified from these hand-to-mouth gestures (HMGs). Rotations or angular motions of the smoking hand during a puff sequence also have distinguishing features. These rotations occur in a certain direction when the hand moves towards the mouth, and in the opposite direction when the hand moves away from the mouth. These rotations can also indicate smoking events [54
Regarding smoke inhalations, smokers generally do not inhale during cigarette lighting [55
], and inhale a very small amount during the initial puffs. To avoid irritation in the throat in the initial puffs, some smokers briefly hold the smoke in their mouth. Major smoke inhalations are done either by deep breathing, and occasionally by ‘Frenching’ (pushing some smoke back into the air without exhaling completely, and inhaling it through the nose—also referred to as a ‘Chinese Drawback’) [55
]. A smoke inhalation can be summarized as a sequential process of: (a) A cessation of normal air-intake (breathing apnea) during cigarette holding; (b) A sharp increase of tidal volume and airflow due to smoke inhalation into the lungs; (c) Occasionally a brief period of smoke holding in the lungs, and; (d) A slow or forced exhalation, either through nose or mouth [56
]. This characteristic respiration pattern may also be an indicator of smoking. Figure 2
illustrates a typical smoke inhalation in terms of changes in the lung breath volume.
There is a significant difference between the acoustic properties of a smoking breath and a non-smoking breath. By characterizing these differences in a non-invasive way, it may be possible to detect the smoking episodes [57
Some instantaneous changes in the physiological parameters of the smoker (such as blood pressure [58
], and heart rate [59
], etc.) also occur during smoking. These parameters, if characterized correctly, may help identify smoke inhalations.
Again, hand-oriented smoking activities (such as cigarette lighting, hand to mouth gestures and a cigarette holding between puffs) may require smokers to frequently look at their hands. An egocentric camera, such as a camera positioned on the head or chest of the person, naturally approximates the visual field of the camera wearer, and offers a valuable perspective to understand the smoking activity and their context in a naturalistic setting.
A total of 51 research studies employing wearable sensors of different modalities addressing these behavioral and physiological manifestations associated with smoking have been reported in the last decade. These approaches have been validated on a number of smoker subjects. Table 3
provides a brief summary of these publications.
This review was intended to provide a systematic evaluation of existing wearable sensors for the objective detection of behavioral and physiological manifestations associated with cigarette smoking. This review identified five specific phenomena related to cigarette smoking that were targeted in the development of wearable sensors. The review also explored 51 research publications describing methods to identify and evaluate smoking-related features assessed through individual sensor systems or their combinations.
The review found evidence that instrumented lighters can capture the initiation of a smoking sequence, and are capable of collecting data on smoking frequency in an unobtrusive way. Further, the lighter can be used in multi-sensor approaches for establishing the beginning of a smoking session. However, if the smoker uses a different lighter than this instrumented one, those particular smoking events will not be detected.
Studies covered in this review suggested that RF Proximity sensors can be effective tools for determining the frequency and duration of hand gestures preceding smoking. In a typical cigarette holding gesture, RF antennas were reported by Sazonov et al. [53
] to produce the highest magnitude of signal strength relative to hand gestures associated with other activities (such as eating). However, these differences in signal amplitude may not be sufficient to differentiate among general hand-to-mouth gestures [63
], but may be capable of providing supportive features to be used for the analysis of smoking patterns in multi-sensor approaches. Furthermore, this approach is typically used to detect gestures of the dominant hand; any smoking using a non-dominant hand (e.g., while driving) will also go undetected. The effectiveness of this method might be limited if a subject generates more frequent hand movements (not related to smoking) near the face. Also, the method will not be able to distinguish whether a person is smoking or resting/reading while supporting his chin with the smoking arm or hand.
IMUs were also found useful for detecting transitions of arm positions during smoking. Initial research with this approach involved the placement of multiple IMUs (3D or 6D) on different hand positions. However, recent research has focused on single IMUs. The 9D IMUs were also employed where concerns of battery longevity were not present. However, IMUs cannot provide information about the absolute position of the arm and its proximity to the mouth. The central challenge of the IMU-based approach is to recognize a smoking gesture ‘in the wild’ without any explicit information from the plethora of other gestures that a user performs each day. Furthermore, there are significant signal variations due to the changes in the users’ body orientation. When people swing their hands during smoking or in conversation, smoking hand gestures are difficult to identify. In some cases, wrist-worn sensors may not remain fixed in the initial placement position, and the sensor responses may vary under free-living conditions. Also, concurrent activities (e.g., walking, talking) while smoking modifies the characteristic pattern of the smoking gesture. Smartwatches also have these inherent limitations; however, they might facilitate real-time intervention (with or without pairing with smartphones).
RIP sensors are effective in capturing variations in the volume of inhaled smoke, the duration of inhalation and breath-holding time, and in bronchial reactivity. The breathing patterns measured through RIP sensors are highly susceptible to artifacts caused by hand and body motions. Stress, speaking, walking or other confounding events also had some effects on respiration measurements [91
]. Processing of RIP sensor signals and robust classification algorithms are required to detect smoking patterns. RIP devices may also be cumbersome if worn for an extended period. Unlike initial implementations, recent RIP approaches contain miniature data logging modules with more comfortable elastic bands. Nevertheless, there are ongoing concerns and issues with clothing integration, cleaning and obtrusiveness of the devices.
Bio-impedance measurement systems are free of the limitations of integration with clothing, however, they require electrode attachment to the body. A combination of Bio-impedance based breathing sensor, ECG sensor and accelerometer placed on the chest was employed by Imtiaz et al. (2019) [111
] for the detection of smoking events. However, the model suffered from many false positives, especially in free-living conditions. This study assumed that changes in heart rate parameters during the study period were either due to cigarette smoking or intense physical activities. Any physiological or ambient factors (with light physical activities) that could lead to a change in heart rate parameters might have caused false positives with this approach. Also, the impact of smoking on physiological signals or the influence of concurrent activities would likely vary greatly among people. Wattal et al. [112
] presented textile electrodes and connectors, which can be evaluated to ease the data collection of the ECG and bioimpedance signals.
Smoking detection based upon acoustic signals is susceptible to ambient noise, hence robust signal processing methods for speech and artifact rejection are necessary for high accuracy. The visibility of this sensor system to others might limit the mass implementation of this approach.
Despite the limitations of wearable sensors and the failure of sensor systems to be 100% accurate in the detection of smoking events in all circumstances, extant systems have identified interesting smoking-oriented phenomena. For example, research using instrumented lighters substantiated the idea that smokers tend to overestimate their smoking consumption, and may be unaware of many instances of their daily smoking [12
]. The instrumented lighter has also identified daily recurrent patterns of smoking incidents on an individual basis [12
]. Work with the breathing sensor has verified that smoking displays a specific breathing pattern [58
]. Further, the combination of breathing and proximity sensors identified individual traits in breathing patterns [102
]. This combination of sensors also demonstrated that anthropometric characteristics (such as obesity, gestures) of the person affect the quality of smoking-specific breathing signals [104
]. Finally, a combination of inertial sensors and instrumented lighters revealed that smokers under surveillance consume cigarettes much faster with a higher number of hand to mouth gestures than when in free-living conditions [108
Additional successes with wearable sensors in smoking research are likely to be achieved if factors such as size and comfort of wearable systems, applicability in daily usage and inconspicuous monitoring are addressed. Due to their form factors, custom wrist-worn inertial sensors or smart-watches might be relatively easier to adopt for daily usage. RIP breathing or acoustic sensors tend to be obtrusive and somewhat cumbersome if worn for an extended period. Hence, the miniaturization and commercialization of these sensor systems will foster their acceptance by all types of smokers.
The camera-based system provides a direct way (from images of lit cigarettes) to detect smoking. Unlike other sensor systems, this sensor could identify the smoking environment, social interaction and locations that may promote smoking. This information related to activity and context during/immediately prior to smoking could play an important role in developing smoking intervention methods. The timestamp embedded in these images can provide additional information on the smoking time of day, duration and frequency. In the feasibility study involving 10 participants, all smoking events were correctly detected; however, it requires further validation, involving many participants of varying age and demographic profiles.
Most of the validation of the above-mentioned systems was limited to laboratory settings. These systems need systematic evaluation under extended free-living conditions. These evaluations need to gather more detailed information about intrusion and comfort. Also, these methodologies need to be tested for significantly longer periods of time (weeks/months) to fully examine their operation before they can be employed for general use.
The number of participant’s involvement varied between validation studies. Out of 51 reviewed articles, 40 studies involved more than 10 smoker subjects (two studies involving instrumented lighter, eight involving RF proximity sensor, 12 involving inertial sensors, 16 involving breathing sensor, one involving acoustic sensor, one involving egocentric camera). However, the remaining 11 studies had tests conducted in very small populations (less than 10, as low as two participants). Since the focus of this review was the presentation of all proposed sensors for smoking detection and characterization, no papers were excluded due to the small study size. However, the strength of the conclusions drawn from such studies is limited.
Individual smoking patterns can be influenced by a variety of factors such as location (e.g., smoking zones, automobiles), ambient conditions, and physical postures (walking, standing, sitting, and relaxing). However, no study on wearable sensors has systematically analyzed the impact of these contextual factors on smoking. Hence, the available mobility sensors in the PACT 2.0 platform (GPS or pedometer) can be evaluated to investigate their impact on improving the accuracy of current smoking detection methods.
Data recording is another important aspect of the available systems. Custom-made sensor systems have either onboard flash storage or the capability of wireless transmission to a nearby receiver or smartphone. In most of these approaches, data can only be accessed offline for computer analysis and cannot assist smokers to react immediately to their smoking situations. The smartwatch based IMU approach [65
], introduced methods to implement real-time detection algorithms at the smartphone to facilitate real-time interventions. The study reported by Skinner et al. [70
] provided an approach to eliminate the necessity of smartphones and integrated everything into a single node (a wristwatch). McClernon and Choudhury [114
] and Qin et al. [115
] proposed methods to use only smartphone sensors (Wi-Fi, GPS, and Accelerometer) data to detect smoking events. These above-mentioned approaches may be capable of recruiting social support groups to inhibit smoking behavior. For more robust interventions, a blend of a Smartwatch platform with other fusion modules could be explored. These systems could even relate to devices on an Internet of Things (IoT) network to develop new intervention strategies.
6. Future Directions
This review demonstrates that the monitoring of cigarette smoking by wearable systems is still in an early stage of development and requires considerable research before it is suitable for general usage. No single sensor system provides a complete and accurate solution for the detection of smoking, the characterization of smoke exposure and other behavioral characteristics of smoking. This systematic review has addressed some successes of wearables in revealing interesting smoking-related phenomena. However, the review has identified a variety of challenges and obstacles to be addressed in future research.
First, no wearable sensor system reached an accuracy of 100% (even in controlled laboratory settings) in the detection of smoking-related features. Existing research targeted all major behavioral and physiological manifestations of cigarette smoking (e.g., lighting, hand gestures); however, body-worn or intraoral chemical sensors could be explored for the detection of smoking and the measurement of smoke exposure. Direct targeting of key chemicals, such as nicotine, may offer a universal approach for monitoring traditional and electronic cigarettes. To improve detection accuracy, further methodological improvements targeting signal processing and pattern recognition should be explored for the sensors currently in use.
Second, very few studies provide quantifiable evidence of user comfort, acceptability and adherence during the studies. These should be assessed in a standardized manner by developing a psychometrically validated questionnaire directed specifically to sensors for monitoring cigarette smoking. Future studies should pay special attention to the objective measurement of adherence, which is critical for the reliability of measurements. Additional sensors may need to be integrated into the wearable sensors, specifically with the purpose of identifying if the wearable is being used.
Third, most wearables have been tested in research settings, and only a few prototypes have been tested for accuracy or applicability under real-life conditions. The huge variability of unscripted human behavior and the impact of a myriad of contextual factors may present significant challenges to some of the sensor systems that test well in the laboratory. Future studies should focus on realistic evaluations under free-living conditions.
Fourth, many of the presented devices operate off-line. The development of real-time detection and notification capabilities may pave the way for the development of sensor-based smoking interventions.