Real-time sun protection decisions in first-degree relatives of melanoma patients

Abstract

Objective

Melanoma is the most serious skin cancer, and consistent use of sun protection is recommended to reduce risk. Yet, sun protection use is generally inconsistent. Understanding the decisional factors driving sun protection choices could aid in intervention development to promote sun protection maintenance.

Methods

In 59 first-degree relatives of melanoma patients, an interactive voice response system (IVRS) employing participants’ cell phones was used to assess twice daily (morning, afternoon) real-time sun protection usage (sunscreen, shade, hats, protective clothing) and decision factors (weather, type of activity, convenience, social support) over a 14-day summer interval where morning and afternoon outdoor exposures were anticipated. Generalized estimating equations and hierarchical linear models were used to examine the effect of demographics and decisional factors on sun protection choices over time.

Results

Sun protection use was inconsistent (e.g., 61% used sunscreen inconsistently). Most strategies were used independently, with the exception of moderate overlap of sunscreen and hat usage. Decision factors were highly relevant for sun protection. For instance, sunscreen use was related to the perception of having adequate time to apply it, whereas shade and hat usage were each related to convenience. Few findings emerged by gender, age, or time of day or year. Significant within-subject variation remained, however.

Conclusions

The findings support continued examination of decision factors in understanding sun protection consistency in real time. Interventions where cues to action and environmental supports work together in varied settings can be developed to improve sun protection maintenance in populations at risk for this common disease.

Introduction

The incidence rate of melanoma, the most deadly form of skin cancer, is on the rise. Melanoma rates doubled between 1982 and 2012, and in 2016, nearly 74,000 new cases are anticipated and nearly 10,000 people are expected to die from it (American Cancer Society, 2016a). Ultraviolet radiation delivered via sunlight is the predominant modifiable cause of melanoma, with approximately 65% to 90% of melanomas caused by sun exposure (Dal, Boldemann, & Lindelof, 2007; Thomas et al., 2007). As such, skin cancer risk reduction involves consistent use of sunscreen and shade-seeking, as well as hats and protective clothing (American Cancer Society, 2016b; Centers for Disease Control and Prevention, 2008). Those at increased risk for melanoma include first-degree relatives (FDRs, i.e., biological parent, sibling, or child; (Ford et al., 1995), or individuals with a sun-sensitive phenotype, such as light skin, red hair, or many moles (Tucker et al., 1997).

Over and above cumulative sun exposure, diverse types of exposure lead to different levels of melanoma risk. For instance, intermittent, recreational sun exposure increases risk more than chronic occupational exposure (Gandini et al., 2005), and even a few sunburns substantially increase risk (Gandini et al., 2005). Yet the predominant self-report assessments for sun protection (Centers for Disease Control and Prevention, 2007) are not designed to assess consistency of sun protection, as they direct respondents to generate a cumulative frequency over a broad time period (e.g., “sometimes” or “most of the time” using sunscreen while outside on a sunny day last summer). Using these global self-report methods, it appears that individuals with a family history of melanoma use sun protection inconsistently (Azzarello, Dessureault, & Jacobsen, 2006; Bishop et al., 2007; Geller et al., 2006), and even well-designed, personalized interventions do not resolve this inconsistency (Azzarello et al., 2006; Glanz et al., 2015; Manne et al., 2010). For example, Glanz and colleagues (Glanz et al., 2015) examined the influence of tailored risk information on sun protection among individuals at risk of melanoma and found that while sunscreen use increased significantly in the intervention group, that substantial inconsistency remained. Understanding the decisions that lead to sun protection use is critical to the development of interventions to address pervasive inconsistency in these behaviors.

Another underexplored element of sun protection involves how individuals may choose between sun protection strategies in diverse situations. The use of multiple sun protection behaviors at the same time is low overall (Bandi, Cokkinides, Weinstock, & Ward, 2010). Yet global self-report assessments for sun protection (Centers for Disease Control and Prevention, 2007) do not reveal how decisions about sun protection may be traded off or used independently (“I’m using sunscreen, so I don’t need protective clothing”), or used conjointly (“I use my hat along with sunscreen”). Recent evidence shows that while shade and protective clothing provide the most efficacious sun protection (Ghiasvand, Lund, Edvardsen, Weiderpass, & Veierod, 2015; Linos et al., 2011), sunscreen may be most heavily utilized (Koch, Pettigrew, Strickland, Slevin, & Minto, 2016) even though it is recommended to be used in concert with other sun protection methods, not on its own (U.S. Department of Health and Human Services, 2014). Further, the efficacy of sunscreen is highly dependent on whether it is applied correctly (Linos et al., 2011). Understanding sun protection with detail and nuance, as well as the decision factors (e.g., perceptions of the weather, social setting, type of outdoor activity) that may dictate specific sun protection choices is an important prerequisite to improving the effectiveness of strategies to improve sun protection consistency through the development of healthy sun protection habits over time (Rothman et al., 2015; Wood & Runger, 2016).

Real-time assessment of sun protection decision making is required for these enhanced understandings. Sun protection research has previously been undertaken via in-depth retrospective interviews and quantitative surveys (Craciun, Schuz, Lippke, & Schwarzer, 2012; Shoveller, Lovato, Young, & Moffat, 2003), as well as paper and pencil diary reports, some of them conducted in real time (Brandberg, Jonell, Broberg, Sjoden, & Rosdahl, 1996; Brandberg, Sjoden, & Rosdahl, 1997; Girgis, Sanson-Fisher, & Watson, 1994; Glanz, Silverio, & Farmer, 1996; O’Riordan, Glanz, Gies, & Elliott, 2008; Yaroch, Reynolds, Buller, Maloy, & Geno, 2006).

The current study employs cell phone assessment in participants’ real world settings, addressing the following Specific Aims. Aim 1: To describe consistency of sun protection practices (sunscreen use, shade-seeking, hat use, clothing use) over time in melanoma FDRs. Use of each behavior over time and how sun protection practices may vary independently versus conjointly over time are considered. Aim 2: To examine the relative importance of decision factors in contributing to diverse sun protection strategies.

Methods

Sample

Melanoma patients (stages I–III) were approached in surgical follow-up clinics in a large, urban cancer center by trained research staff as per surgeon’s approval. Patients were informed about the study, and were asked to refer any of their potentially eligible FDRs for study participation. To be eligible, FDRs needed to be age 18 or over, English speaking, report at least 1 hour of outdoor activities every morning and afternoon, to be the only participating FDR for a patient, and to endorse that they use sun protection at least sometimes. We set inclusion criteria for sun exposure (at least 1 hour of anticipated sun exposure every morning and afternoon) and sun protection (at least some use of sun protection) in order to allow for the relevance of active decision making around these behaviors during the morning and afternoon time periods.

The study was approved by the Memorial Sloan Kettering Cancer Center Institutional Review Board (IRB). Informed consent procedures were conducted either in clinic, if FDRs were attending clinic with a patient, or verbally by telephone.

Procedure

An interactive voice response system (IVRS) was developed and programmed to reach participants on their phones. Any reliable phone could be used, either cell phone or landline, as chosen by the participant. Responses were recorded through telephone keystrokes. The IVRS generated a telephone call twice daily to participants regarding their sun protection choices and decision factors at 12:30 pm (to assess morning sun protection and decision making) and 5 pm (to assess afternoon sun protection and decision making) across a 14-day period from June to October. Accordingly, “real time” assessment was operationalized as twice daily assessment. This is consistent with a time-based approach to ecological momentary assessment (Shiffman, Stone, & Hufford, 2008) because it was anticipated that participants’ decisions about sun protection could be easily conceptualized discretely in the morning and afternoon, and that more frequent assessments (4–6 times daily) would risk poorer adherence without corresponding value for more detailed sun protection decision making. They also completed an audio narrative to provide additional context surrounding their sun protection use at each assessment, and these qualitative findings are reported separately (Fitzpatrick et al., 2016). Participants elected the 14-day summer assessment period of their choice.

Measures

The IVRS used a survey specifically developed for melanoma FDRs (see Figure 1) using ethnographic decision tree methods (Beck, 2005; Gladwin, 1989) drawn from an ethnographic research tradition (Spradley, 1980), which is a qualitative approach to assessing decision making processes in real-world settings. This methodology has been used in anthropology and psychology for 25 years to model medical treatment decision making (Weller, Ruebush, & Klein, 1997; Young, 1980), needle-sharing decisions among drug users (Johnson & Williams, 1993), and decisions to recycle beverage cans (Ryan & Bernard, 2006). Ethnographic decision tree modeling was used in a sample of 25 melanoma FDRs; 21 common decision factors related to four sun protection behaviors (sunscreen use of at least SPF 15, shade-seeking, use of hats and sun protective clothing) in melanoma FDRs were identified (Shuk et al., 2012). The in-home interview is the recommended strategy for EDTM (Gladwin, 1989) as it allows the interviewer to obtain a first-hand sense of the contexts within which sun protection is performed, since participants interacted directly with their sun protection items during the interview. The interview focused on recall of two separate recent sun exposure periods (varying by setting and activity) when they were outdoors for one hour or more. Participants were asked to report on their use (or non-use) of the four sun protection methods under examination for each sun exposure period, as well as contextual background, including the outdoor activity and setting, other individuals present, weather conditions, time of day and length of time when outdoors. This strategy on reporting behavior and decision-making for two contrasting episodes is a central feature of EDTM that highlights issues of behavioral inconsistency across different settings. For data analysis, we constructed four decision tree models for each participant. Each decision tree model (one for each sun protection behavior) depicted a series of ordered, discrete decision factors, both facilitators and barriers, followed by yes/no choice points. For details on EDTM data analysis see Shuk and colleagues (Shuk et al., 2012).

Figure 1.

Interactive Voice Response System (IVRS) Survey. Includes four behavioral questions and 21 decision factor questions. The abbreviated name for each decision factor is included in italics.

At each morning and afternoon assessment over 14 days, the survey evaluated use (yes/no) of each of the four sun protection behaviors (sunscreen use of at least SPF 15, shade-seeking, hat and protective clothing use) and whether each of the 21 decision factors identified in previous research (Shuk et al., 2012) were relevant in influencing sun protection use in that particular time assessment (yes/no). Demographics including sex, age, race/ethnicity, educational attainment, marital status, employment status, personal history of melanoma, number of family members with melanoma, and whether participants were on vacation during the study assessment period were assessed by telephone prior to the beginning of the 14-day assessment period. Finally, after the 14-day assessment period, participants were reached by telephone to complete a study satisfaction survey; findings are reported separately (Holland et al., 2017).

Statistical Approach

To examine Aim 1, descriptive statistics were calculated to summarize use of sun protection practices over the 14-day assessment period, morning and afternoon, for a maximum of 28 observations per person per sun protection strategy. Frequencies and variability for each of the four sun protection behaviors were examined. Environmental contexts were observed rather than manipulated, so a crude method of identifying consistent users was developed such that use of any sun protection strategy more than 80% of the time was defined as consistent use; use less than 10% of the time was defined as rare and use between 10% and 80% defined as inconsistent. To assess whether sun protection is used conjointly, the correlation of sun protection strategies using a pairwise gamma statistic to assess co-incidence in binary (yes/no) sun protection behaviors was calculated. Gamma for ranks is more suitable than Cohen’s Kappa as it allows for negative values for inversely-correlated patterns, where a value of −1 indicates perfectly inverse correlation and values of zero and +1 indicate, respectively, no association and perfect correlation, with smaller values indicating the degree of association. Gammas were calculated for each of the 28 time periods of the study, with the standard error over these 28 intervals used to calculate a p-value.

To examine Aim 2, we first assessed which decision factors were associated with sun protection behaviors. For those that were associated with behaviors, we assessed the sources of variation, either between-person or within-person. Generalized estimating equation (GEE) models were employed to assess the relationship between each sun protection behavior and each decision factor. In each logit model, the outcome was a binary indicator of one of four sun protection behaviors (used/did not use sun protection), the predictor was a binary response to a decision factor question (yes/no), and an autoregressive correlation structure was used. Odds ratios and confidence limits were derived from each predictor’s estimated beta coefficient and its limits, and provide a way to compare the influence of each decision factor. The alpha-level was conservatively set a priori to 0.002 to account for multiple hypothesis testing. To validate the unadjusted associations in the single-predictor models, for each sun protection behavior a multivariable model was run that included as predictors all of the decision factors significantly associated with the given behavior in the unadjusted (crude) models. In this GEE analysis separate adjusted models were fitted for each sun protection behavior, ignoring the effects and correlation of other sun protection behaviors.

We further investigated the impact of within-subject variation beyond between-subject variation via an extension of Ecological Momentary Assessment methods (Schwartz & Stone, 1998). These authors illustrated the statistical methods in examining moment-to-moment measurements of negative affect as a function of between-person and within-person differences. For the between-person differences, these authors examined whether or not people with lower socioeconomic status (SES) would report more negative affect compared to those with higher SES; for the within-person differences, they postulated that participants would report elevated negative affect on stressful days compared to average days. The within-person stress is quantified by centering momentary stress by each participant’s average stress level. We used a similar modeling framework. Each participant’s average on a decision factor (e.g., percentage of ‘sunny and hot’ days over 14 day assessment) represents the between-person decision factor, and it was hypothesized that those who reported 80% hot and sunny days would use sunscreen more frequently than those who reported 20% hot and sunny days. The within-person differences were operationalized as the daily hot and sunny reports centered on each participant’s 14-day average. This two-level model form was fitted for the ith participant at the jth timepoint, given decision factor (df) for the time interval:

$$\mathit{logit}\left({\text{Behavior}}_{ij}\right)={\alpha }_i+{\beta }_0·\phi \left(\overline{{\text{df}}_{\mathrm{l}}}\right)+{\beta }_1·\left[\phi \left({\text{df}}_{ij}\right)-\phi \left(\overline{{\text{df}}_{\mathrm{l}}}\right)\right]+{\beta }_2·X_{ij}+{\epsilon }_{ij}$$

where α_i = α₀ + α₁ · W_i + δ_i, and φ indicates the arcsin transformation. Between-person differences are tested by the β₀ parameter; within-person differences are tested via the β₁ parameter. A random intercept was also included, and potential confounders were included for adjustment at either level-1 (i.e., days since solstice, time of day) or level-2 (i.e., age, gender) via the X_ij or W_i term, respectively. Arcsin was deemed most appropriate for transforming the binomial predictors to a near-normal approximation due to an abundance of values at the extremes of the distribution (Cohen, 1988). All statistical analyses were conducted in R, version 2.15.2, or SAS software, version 9.2, and missing data was handled by casewise deletion.

Results

A total of 512 patients were approached over the course of three summers (2011–2013, June through October) and almost half of the patients approached (n=251, 49%) referred at least one FDR. In total, 418 FDRs were referred. About half (n=214, 51%) were found to be ineligible, and this was predominantly (n=173, 81% of ineligibles) due to a lack of daily outdoor activities. Eligibility was not assessed in 123 FDRs (30%) due to study refusal or our inability to reach them by telephone. Of 81 eligible FDRs, 69 (85%) consented to the study. Of the 69 participants who consented, one participant withdrew before providing data, and nine withdrew after only providing demographic data. Fifty nine participants (86%) completed at least one telephone call and thus comprised the study sample; of this number, 53 participants (77%) completed the study in its entirety and study satisfaction was quite high (Holland et al., 2017).

The study participants (N=59) included 22 males and 37 females, and on average participants were 48 years of age (range 18–82 years). Additional baseline characteristics are reported in Table 1. Data included responses to between 1 and 28 IVRS surveys (morning and/or afternoon) per participant, for a total of 1,312 records. Out of a possible 1,652 IVRS surveys (59 participants * 28 observations), most (79%) data was complete (1,312 surveys).

Table 1.

Participant demographics (N=58)^*

Demographic	n	%
Sex
Male	21	36
Female	37	64
Age
Under 50 years	30	52
50 years or more	28	48
Race/Ethnicity
White/Hispanic	1	2
White/Non-Hispanic	57	98
Education
High school graduate	6	10
Some college	9	16
College graduate	22	38
Graduate degree	21	36
Marital status
Married	36	62
Divorced/Widowed	5	9
Single	15	26
Unmarried couple	2	3
Current employment status
Employed	29	50
Self-employed	9	16
Student	4	7
Homemaker	1	2
Retired	12	21
Unemployed	3	5
Personal history of skin cancer
No	52	90
Yes	6	10
No. of family members with a melanoma
1	45	78
2	8	14
3	4	7
Missing/Unknown	1	2
Planning to be on vacation during study
No	41	71
Yes	17	29

Note:

^*All non-gender demographic information was missing for one additional male

Consistency of sun protection practices (Aim 1)

See Table 2 for Aim 1 findings. Sunscreen use was reported most frequently, with 49% of all responses (n=636 out of 1,312 observations) indicating use. All four sun protection strategies were used quite often, however, with 47% (n=612) of responses indicating shade-seeking, 36% (n=465) indicating hat use, and 43% (n=541) indicating protective clothing use. As expected, many participants used each sun protection strategy inconsistently over the 14-day study period. Among the 56 participants who answered the IVRS survey at least twice, a majority (n=34, 61%) reported inconsistent sunscreen use, defined as using sunscreen 10% to 80% of the time. Almost one fifth (n=10, 18%) used sunscreen rarely (i.e., during fewer than 10% of their reporting periods). Similar findings were evident for protective clothing with more than half (n=32, 57%) giving inconsistent responses, and one quarter (n=14, 25%) responding “Yes” to fewer than 10% of their reporting periods. For shade-seeking and hat use as well, most participants (75% and 63%, respectively) reported inconsistent use.

Table 2.

Variability in four sun protection behaviors, across time and participants (N=59)

Source	Measure	Sunscreen	Shade-Seeking	Hat use	Protective Clothing
Across observations, percentage	Behavior prevalence	48.8%	47.2%	36.0%	42.7%
Across participants, percentage	Rare use (<10%)	17.9%	8.9%	23.2%	25.0%
	Inconsistent use (10% to 80%)	60.7%	75.0%	62.5%	57.1%
	Consistent use (>80%)	21.4%	16.1%	14.3%	17.9%
Across time, gamma (p-value)	Sunscreen	----	−0.03 (0.50)	0.39 (<.0001)	−0.11 (0.01)
	Shade-seeking		----	0.19 (<.0001)	0.03 (0.48)
	Hat use			----	0.01 (0.87)

Note: Prevalence is calculated as the percentage of all observations in which the behavior was used. Use is calculated for each participant (excluding those with only one observation), for each behavior, percentages represent the distribution of participants falling into each use category. Pairwise gammas for co-occurrence are first calculated for each of 28 timepoints over duration of the study, and then an average taken for each pair.

Overall, the gamma findings did not reflect universally high pairwise associations between the sun protection behaviors. The highest was a significant, moderately sized association between sunscreen and hat use averaged across the 28 gammas (mean = 0.39, p < .0001), indicating that sunscreen and hats were used together about 40% of the time. There were smaller associations between use of shade and hats (mean = 0.19, p < .0001), and sunscreen and protective clothing (mean = −0.11, p = .01). The association between sunscreen and protective clothing was negative, indicating that these behaviors sometimes precluded each other.

Relative importance of decision factors in sun protection (Aim 2)

Sunscreen use

Participants were more likely to use sunscreen when they had time to apply it (OR=3.2, CI: 2.0–5.1) and when they were encouraged to use sunscreen by others in their lives (OR=3.1, CI: 1.8–5.4). Other significant factors associated with sunscreen use in the univariable models included having sun in one’s eyes or face, sunny and hot weather, being outdoors for a longer period of time, being at a water setting, being in the sun for a longer period of time, being in the sun during peak times, being engaged in physical activity, and feeling hot and uncomfortable (all ORs=1.9–2.8). Participants were less likely to use sunscreen when it was cloudy (OR=0.5, CI: 0.4–0.7). In the multivariable model, Figure 2, all but three of the variables remained significant. The strongest factor associated with sunscreen use remained having time to apply it where the magnitude of effect was undiminished (adjusted OR=3.2, CI: 1.9, 5.6).

Figure 2.

Odds ratios with confidence intervals for unadjusted associations of decision factors on sun protection behaviors. Significant associations are present when the confidence interval does not cross the OR=1 reference line. Significant (p< 0.002) predictors from unadjusted models that remained significant in adjusted models are shown in red and with an asterisk. Decision factor names are abbreviated from the full survey.

Shade-seeking

Participants were more likely to seek shade when it was sunny and hot outside (OR=2.4, CI: 1.8–3.2), when their skin started to hurt or burn (OR=2.4, CI: 1.5–3.9), and when they felt hot and uncomfortable (OR=2.3, CI: 1.7–3.0). Other decision factors significant in the univariable models included shade being conveniently available, and it being too hot outside for clothing (ORs=1.9–2.1). Participants were less likely to seek shade when it was cool (OR=0.6, CI: 0.5–0.8) or cloudy outside (all ps<.05). In the multivariable model in Figure 2, the most important factor was shade being conveniently available (OR=2.6, CI: 2.0–3.5), and it being sunny and hot outside (OR=1.8, CI: 1.3–2.6).

Hat usage

Decision factors associated with hat use included having a hat conveniently available (OR=13.3, CI: 8.5–20.9), and being at a water setting (OR=2.5, CI: 1.5–4.1). Other significant factors included being in the sun or outdoors for a longer period of time, sun protection encouragement from others, engagement in or watching a physical activity, being in the sun during peak times, and feeling hot or uncomfortable in the sun (OR=2.3, CI: 1.7–3.0). In the multivariable model in Figure 2, the decision factor most strongly associated with hat usage remained having a hat conveniently available (OR=18.1, CI: 10.3–32.0), and being in the sun during peak times also remained significant (OR=1.7, CI: 1.3–2.4).

Protective clothing usage

The three decision factors that were associated with protective clothing in the unadjusted models were those having to do with the weather. Participants were more likely to wear protective clothing when it was cool (OR=2.8, CI: 2.0, 3.8) and when it was cloudy (OR=1.5, CI: 1.2, 1.9). Participants decided against clothing use when they were feeling hot or uncomfortable in the sun (OR=0.6, CI: 0.5–0.8). In the multivariable model in Figure 2, cool weather remained associated with clothing use (OR=2.5, CI: 1.8–3.5).

Models separating out within-subject variation from between-subject variation were fitted for the two most significant predictors found for each behavior, see Table 3. Overall, the within-subject variation parameters were comparable to the between-subject variation parameters, though only the within-subject parameters were consistently significantly associated with sun protection behaviors (all ps ≤.01). In the Supplementary Figure, participants’ proportion of days with a given decision factor are plotted against the proportion of days with the specific sun protection behavior; slopes far from zero indicate large between-subject effects, while large mean differences indicate within-subject effects. The fit lines for proportion of days that sunscreen is used are relatively flat, which indicates that any association between the decision factor “sunny/hot” day and sunscreen use is not due to between-person variability. In fact, there is substantial within-subject variation, as indicated by the mean differences, on days that were perceived and not perceived to be sunny and hot. In contrast, the slopes for “time for sunscreen” and “no time for sunscreen” clearly differ, indicating between-person variation is important in the association. We re-ran these models examining potentially relevant covariates, including time since summer solstice as a proxy for time of year and time of day [morning/afternoon], age and gender. Though a few were significantly associated with sun protection behaviors (e.g., age with hat use; results available from the primary author), none of these covariates changed the results for decision factors discussed above.

Table 3.

Between-participant and Within-participant effects of select Decision Factors on Sun Protection Behaviors (N=59)

Sun Protection Behavior	Decision Factor	Between-Participant Parameter (p-value)	Within-Participant Parameter (p-value)
Sunscreen	Sunny/hot	1.64 (0.12)	1.25 (<.0001)
Sunscreen	Time for sunscreen	1.94 (<.01)	1.23 (<.0001)
Shade	Sunny/hot	2.29 (<.01)	0.71 (<.0001)
Shade	Shade available	1.47(<.01)	0.65 (<.0001)
Hat	Hat available	3.48 (<.0001)	1.93 (<.0001)
Hat	Peak sun	1.13(0.09)	0.77 (<.0001)
Protective clothing	Cool outside	0.48 (0.32)	0.74 (<.0001)
Protective clothing	Sunny/hot	−0.97 (0.33)	−0.28 (0.01)

Note: Results are based on HLM models regressing the given sun protection behavior on: the arcsin-transformed mean of the longitudinal indicators for the decision factor (i.e., between-participant effect), the mean-centered arcsin-transformed indicators at each timepoint (i.e., within-participant effect), and a random intercept per participant.

Discussion

Given that even a few lifetime sunburns can substantially increase risk (Gandini et al., 2005), consistent sun protection (sunscreen, shade, hats, and protective clothing) is exceedingly important to melanoma risk reduction efforts. Yet even in higher risk individuals, sun protection is practiced inconsistently. Measurement of sun protection choices, and the decision factors underlying these choices, is difficult to assess globally and retrospectively, so a real-time assessment strategy using an IVRS was developed. This paper is consistent with the goal to conduct research in real-world, real-time settings (Mermelstein & Revenson, 2013; Spruijt-Metz et al., 2015) and specifically to examine contextual, multi-level factors in health behavior maintenance over time. Our findings indicate, not surprisingly, that even in this sample of melanoma FDRs of whom 30% were on vacation, sun protection usage was highly variable. This was true even though participants were not eligible for the study unless they reported at least 1 hour of outdoor summer exposure each morning and afternoon. Across the 14-day study period, most participants gave inconsistent responses – sometimes using and sometimes not using – for each of the four sun protection strategies. Of note, the use of sun protection showed few distinct differences across time of day, gender, and age of participants.

Many of the decision factors generated in prior qualitative work (Shuk et al., 2012) were quantitatively associated with sun protection use in the current study. The decision to use sunscreen was related to the weather, having social support for sunscreen use, and the perception of having adequate time to apply it. Additionally, outdoor exposure was an important factor in making a decision to use sun protection, such as feeling hot and uncomfortable, being at water settings, or engaging in or watching sports outdoors. Such contexts serve as cues to action for sunscreen use. Primary barriers to sunscreen use were also weather-related, including cloudiness or less intense sun. While there was some similarity in decisions for sunscreen use and shade-seeking – sunny, hot weather promoted shade-seeking as well as sunscreen use – there were also significant contrasts. The availability of shade was a strong promoter of shade use, which supports the importance of environmental interventions to provide shade options. Hat usage was largely driven by convenience, and also by engagement in sports or a physical activity, as well as being in the sun during peak exposure times. Decisions to use protective clothing were largely driven by cooler temperatures, and feeling hot or uncomfortable presented a barrier to use. These findings provide a potential context and rationale for why sun protection strategies are largely independent - these are different behaviors made for different reasons. Shade use may be adopted for short intervals when it is available, while sunscreen, shade, and hats may be adopted during the sunniest weather. Finally, in cool, cloudy weather shade-seeking was less likely, where some sunscreen and hats may be used. These findings could be integrated into adaptable interventions to promote sun protection consistency in different settings, such as outdoor exercise or when social support is unavailable (Nahum-Shani, Hekler, & Spruijt-Metz, 2015).

Furthermore, there remained substantial within-subject variation for all sun protection strategies examined. These findings indicate some intriguing areas for future descriptive and methodological work, including examination of a wider range of decisions that may dictate diverse sun protection choices over time, explaining more of the within-subject variation, as well as opportunities to more precisely measure habitual sun protection usage. Additionally, the extent of the within-subject variation found here provides some direction for the development of precision targeting approaches to intervention messages that address the potential sources of the great variation in sun protection usage we found in this study, such as diverse contexts, outdoor activity types, and preference for sun protection type.

There are also measurement implications of the study findings. First, there was some support that use of certain sun protection strategies were related to each other, with the strongest support being the combined use of sunscreen and hats. Yet, there is not strong justification for any underlying latent “sun protection” factor in this longitudinal study; those who are forgoing one form of sun protection (such as shade-seeking) are not necessarily more likely to practice another (such as sunscreen). This argues for continued examination of diverse sun protection strategies as independent behaviors. Secondly, we support the continued assessment of decisions regarding sun protection in real-time. Participants were not burdened by, and were adherent to, the IVRS (Holland et al., 2017). Further, the findings comparing within-subject to between-subject variation indicate that within-subject differences in “momentary” decision factors were by far more important than the more static, between-subject differences.

Finally, many studies usefully examine individual-level demographic and attitudinal factors in sun protection uptake and maintenance (Kasparian, McLoone, & Meiser, 2009) that have been derived from health behavior theory (Conner & Norman, 2005). Theoretical models based on individual-level covariates are essentially static and thus limited in their capacity in explaining complex dynamics over time. Our prior qualitative work (Shuk et al., 2012) guided the distinctly different approach in this study. Future work should examine inter- and intra-individual attitudinal predictors of sun protection along with decisional factors. Further, our findings of the real time audio diaries associated with the IVRS revealed additional detail regarding sun protection decisions that could be further examined, including emotional reactions to sun protection decisions, such as guilt and regret (Fitzpatrick et al., 2016). Examination of the distinct environmental settings and predictors for uptake and maintenance of various sun protection strategies is warranted. While most sun protection strategies were not used independently, it could be that joint sun protection, such as use of sunscreen and hats together, may be more common in some environmental contexts. These important questions are beyond the scope of this work, but the importance of decisions around sun protection dictates inclusion of such factors in future work on both those at elevated melanoma risk as well as the general population.

Study limitations include the fact that only those who engaged in some sun protection were eligible; findings cannot be generalized to decision making in those melanoma FDRs who did not report regular morning and afternoon sun exposure, or never use sun protection. Further, decision making in those who were on vacation during the study cannot be generalized to non-vacation behavior. The findings are based on self-report, which is potentially sensitive to social desirable reporting. While these self-reports were elicited twice daily, a number of hours may have elapsed between sun protection decisions and the next survey. We did not collect weather information, which could have provided useful decision validation, and we did not specify type of hat used for sun protection. Finally, while study adherence was high, continued examination of new technologies for real time data collection - such as text messaging - is warranted.

In conclusion, these findings indicate a series of relevant decision factors associated with sun protection use, a usable IVRS to assess real-time sun protection, and direction forward to understand the high inconsistency in sun protection use in individuals at risk for melanoma. The findings argue for a multi-level, contextual approach to sun protection where cues to action and environmental supports work together with attitudinal factors to promote sun protection maintenance over time in populations at risk for this deadly, but common, disease.