Participants

Three hundred ten participants were recruited from Rutgers University via email listservs, social media platforms, flyers, and word-of-mouth. Eligibility was determined based on a brief online survey accessed using a link or QR code. Participants were deemed eligible if they were female with a regular menstrual cycle between 25–35 days, enrolled as a full-time university student, 18–25 years old, had no history of psychosis, were not pregnant or breastfeeding, and had no current physical conditions that required daily medication (except for hormonal contraception (HC)). Rutgers University’s Institutional Review Board approved this study for the protection of human subjects involved in research. Sample characteristics are shown in Table 1.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Sample characteristics.

https://doi.org/10.1371/journal.pone.0353167.t001

All participants provided digital informed consent prior to completing the screener and survey. Eligible participants were invited via an individualized emailed link to a Qualtrics survey. Email reminders were sent to prompt initiation and completion of the survey. Ten participants were excluded from analysis for either finishing <25% of the survey or for inputting the same answer for >75% of survey items; thus, the final sample was 300.

Survey measures and design

The current survey included (a) demographics, health, and behavior surveys to assess participant characteristics; (b) supplemental validated surveys to illustrate factor qualities and ground findings in extant literature, and (c) a list of “somatic items”, which were words or brief phrases that depicted bodily states, physical perceptions, and sensory experiences, to assess the frequency and valence of somatic experiences. The identification, selection, and refinement of these items is presented in Results.

The survey used for the present analyses consisted of 4 blocks. Block 1 was first for all participants and included demographics, health history, reproductive health, exercise, and alcohol use questions. Basic demographics collected included age, year of university study, and race and ethnicity. A modified Reproductive Status Questionnaire [7] was used to categorize women who used HC and those who were naturally cycling. Exercise was assessed as frequency (typical days per week) and duration (minutes per day, in intervals: < 10 minutes, 10–20 minutes, 21–30 minutes, 31–40 minutes, 41–50 minutes, 51–60 minutes, and >60 minutes) of aerobic and resistance exercise, with an option to indicate “No exercise”. For the analyses, duration intervals were converted to numbers at the midpoint of each interval (<10 min = 5 min, 10–20 min = 15 min, 20–30 min = 25 min… > 60 min = 65 min). Final exercise variables included exercise status (i.e., exerciser or non-exerciser); aerobic exercise status (yes or no) and weekly exposure (minutes per week); and resistance exercise status (yes or no) and weekly exposure (minutes per week). Alcohol use was assessed as the last time they consumed more than a sip of alcohol (within the last week, 1–2 weeks ago, 3–4 weeks ago, more than a month but less than 3 months ago, more than 3 months ago, and never) to categorize “non-drinkers” (did not consume alcohol in the past 4 weeks) and drinkers (did consume alcohol in the past 4 weeks). Drinkers reported the number of standard drinks consumed and the number of hours drinking every day for the past 30 days using an online version of the Timeline FollowBack (TLFB; SMASH Labs, Providence, RI) [33]. Alcohol variables included past 30-day total frequency and quantity and the maximum number of drinks consumed in a single drinking day.

Blocks 2–4 each contained supplemental items from validated surveys and 50–70 newly identified somatic items (described below). Fourteen supplemental survey items were from the first section of the Premenstrual Symptoms Screening Tool (PSST) [29]. A sum score of all 14 items (“Not at all” (1), “Mild” (2), “Moderate (3), and “Severe” (4)) was created as a proxy measure of premenstrual symptom severity. The items showed excellent internal consistencies (α = .91). Higher PSST scores reflect more severe premenstrual symptoms. Eighteen supplemental survey items were from the Body Awareness Questionnaire (BAQ) [15]. Items were scored on a 7-point scale, from 1, “Not at all true of me” to 7, “Very true of me” and summed to create a total score of body awareness. The items showed strong internal consistencies (α = .87). Higher BAQ scores reflect more awareness of bodily processes and comfort with normal bodily functioning. In addition, participants were presented with a total of 204 somatic items distributed across Blocks 2–4. They rated the frequency that they experienced each somatic item in the past 30 days on a 5-point Likert scale: “Never” (0), “Rarely” (1), “Sometimes” (2), “Most of the time” (3), or “Always” (4). On the next screen, participants were asked to provide valence ratings on a 7-point Likert scale, from “Very unpleasant” (−3) to “Very pleasant” (+3), with 0 being “Neither unpleasant nor pleasant”. Participants also had the option to rate the item as “Unclear” (not scored) if they could not imagine what an item would feel like or if the meaning of the item was unclear; these items were not rated for valence. Items endorsed as “Never” (in past 30 days) were included in valence ratings as they might have been experienced outside of the past month (e.g., rare or acute events, such as being sick).

Blocks 2–4 were counterbalanced across participants, and somatic items within blocks were randomized. Prior to each block, a brief introduction was provided that reminded participants that the somatic items were describing physical sensations, that they should answer based on their own personal physical experiences of their body, and that there were no right or wrong answers. In each block, subjective experience of each somatic item was assessed across two domains. All participants rated 1) frequency and 2) valence of each somatic item. Participants were compensated with an Amazon gift card worth $5 for each block of the survey they completed, up to $20.

Data analysis

Somatic item identification and refinement

An extensive literature review was completed to identify somatic states related to general physical health as well as those related to menstrual health and substance use. The review leveraged existing survey banks (e.g., NIH Toolbox, HealthMeasures, PROMIS), PubMed and GoogleScholar searches, and general internet searches of somatic descriptors (e.g., words or phrases used in somatic therapy or Eastern medicinal techniques). Search terms included words and phrases, such as “somatic”, “physical sensations”, “bodily experiences”, “intoxication”, “menstrual symptoms”, “hangover”, “interoception”, “proprioception”, and “body awareness”. Somatic sensations across a wide number of body systems and physiological experiences that would, theoretically, rate along a full valence spectrum from very unpleasant to very pleasant, were identified. An initial list of 300 items was compiled (see S1 Appendix for full list).

Variable reduction and selection were iterative using both data- and theory-driven processes. First, a pre-data conceptual item reduction step culled the original 300 items to 204 by removing synonymous terms and items with multiple/unclear interpretations. This step was supported by 5 female undergraduate research assistants in the laboratory who were not involved in the item selection for or the development of the survey. These individuals were presented with a list of items and asked to identify, from their perspective, highly synonymous terms, confusing terminology (i.e., words they were unfamiliar with or words that had multiple/unclear interpretations), and terms they considered scientific or medical jargon.

A post-data conceptual item reduction step was undertaken to align items with the initial goals of the project: to create a comprehensive measure that covered multiple body systems and somatic experiences of young women, with emphasis on the ability to measure changes associated with the menstrual cycle and/or behaviors, such as alcohol use or exercise. Low endorsement items and unclear items were removed using frequency ratings from the total sample, and from the non-drinker and drinker subsamples (because alcohol use can create specific somatic experiences (e.g., spinning, high, lips numb, body rush)). Twenty items were trimmed based on ≥20% of both subsamples rating the item as “Unclear”. Four words were trimmed based on being reported as “Never” by 61% of the full sample. Five duplicates were removed. This initial variable reduction concluded with 175 somatic items for imputation and further analyses.

Mean valence was then computed for each item using data from the total sample. Items with an average valence <0 (n = 120) were included in the “unpleasant” dataset; items with an average valence >0 (n = 55) were included in the “pleasant” dataset. This parallels the Positive and Negative Affect Schedule [40] and ensured that both unpleasant and pleasant items were maintained.

Item clustering and validation

Unpleasant items.

A data-driven item reduction step was completed via several iterations of PROC VARCLUS on the 120 unpleasant items, which allowed observation of clustering patterns and identification of items that consistently fit poorly within a cluster. Cluster number was maximized at a 53-cluster solution. The goal was to allow the procedure to quantitatively group items that could be considered synonymous into their own clusters to eliminate items that were very close in meaning and interpretation. Each cluster consisted of 1–4 items and were examined for synonyms and themes. When a synonym was identified (e.g., nauseous, nauseated), the synonymous item that had the lower percent of “Unclear” ratings or greater percentage of endorsement above “Never” was maintained. Thirty items were dropped during this stage. Additional iterations of PROC VARCLUS were performed to identify items that did not load well onto clusters or were frequently in their own cluster; items were removed (n = 10) unless they were considered as critical to the survey’s purpose (e.g., tender breasts). Iterations continued until quantitative fit (R² > .5) and theoretical interpretation were maximized. The final first-order model of unpleasant items consisted of 30 first-order clusters of 80 items.

Second-order clustering of unpleasant items was then performed to tap into dimensional grouping of first-order clusters to be considered as latent factors. The selected model identified 14 second-order clusters. Five second-order clusters contained only one first-order cluster and were not considered in the factors, which eliminated 8 items; the single cluster containing “tender breasts” was maintained. A post hoc conceptual refinement step eliminated 19 additional items for the following reasons: redundancy, having emotional or cognitive interpretations, or misalignment with conceptual factor themes. For redundant items, the original frequency data were referenced and the items with the least percentage of unclear ratings and greatest endorsements above “never” were kept; 5 redundant items were removed. Items determined to have substantive emotional or cognitive interpretations (n = 6) were culled. Finally, after identification of the unifying theme and naming of each second-order cluster as a factor, items that did not seem to align with the conceptual theme of its second-order cluster were also removed (n = 8).

The best fitting second-order model for unpleasant items included 9 second-order clusters that included 53 items (Table 2). The clusters were examined for overarching themes and named based on the unifying theme across items: Sympathetic Activation, Generalized Pain, Gastrointestinal (GI) Distress, Illness, Kinesthetic Disconnect, Low Energy, Neurological Discomfort, Metabolic Stress, and Heat & Urgency Response. Sum scores from unpleasant clusters were all significantly correlated with one another, with correlation coefficients ranging from 0.25–0.68 (S1 Appendix, Table 1). Illness had the weakest correlations with other unpleasant clusters (r = 0.25–0.41).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Unpleasant item cluster identification and factor validation.

https://doi.org/10.1371/journal.pone.0353167.t002

A CFA was then performed and unpleasant items with modification indices > 15 were removed. These included “heart palpitations”, “joint pain”, “abdominal cramping”, and “restless”. The final 9-latent factor model for the 49 unpleasant items (S2 Appendix) showed good model fit based on RMSEA and SRMR but CFI and TLI were below the threshold for a “good” fit (S1 Appendix, Table 2); all item loadings were significant (Table 2), and all latent factors were intercorrelated. The latent factors are illustrated in Fig 1.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Graphic of the identified pleasant (in blue; Physical Alignment, Physical Intimacy, Activated & Attractive, and Parasympathetic Activation) and unpleasant (in red; Sympathetic Activation, Generalized Pain, Gastrointestinal Distress, Illness, Kinesthetic Disconnect, Low Energy, Neurological Discomfort, Metabolic Stress, and Heat & Urgency Response) latent factors.

https://doi.org/10.1371/journal.pone.0353167.g001

Pleasant items.

Fewer positively-valenced items were identified from the literature and included in the survey. Thus, item reduction of the 55 pleasant items was more conservative than with unpleasant items to retain as many items as possible. Criteria for removal during the data-driven item reduction steps were similar to unpleasant items regarding synonyms and loadings; 17 items were removed during this stage.

The first-order cluster solution contained 18 clusters of 39 items. The correlation matrix from these 18 clusters were input into PROC VARCLUS to obtain a final model of 5 second- order clusters (Table 3); one single-item first-order cluster (smooth skin) was excluded. Like the final unpleasant clusters, the final 4 pleasant clusters were subjected to a post hoc conceptual refinement step to address item redundancy, non-physical interpretation, and conceptual fit. For redundant items, the item with the lowest “unclear” rating or the highest non-never frequency rating was maintained; 7 were removed. Finally, after identifying the unifying theme and naming each second-order cluster as a latent factor, items that did not seem to align with the theme of its second-order cluster were also removed (n = 3).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Pleasant item cluster identification and factor validation.

https://doi.org/10.1371/journal.pone.0353167.t003

The best-fitting model for the pleasant items identified 4 second-order clusters of 28 items named based on conceptual similarities: Physical Alignment, Physical Intimacy, Activated & Attractive, and Parasympathetic Activation. Sum scores of pleasant clusters were all significantly correlated with one another (r = 0.40–0.64) (S1 Appendix, Table 3). Most of the correlations were moderate to strong.

A CFA was then performed and pleasant items with modification indices > 15 were removed. This included 4 items: “not bloated”, “smooth digestion”, “strong”, and “well groomed”. The final four-latent factor model for 24 pleasant items (S2 Appendix) demonstrated acceptable model fit based on RMSEA, SRMR, and CFI; TLI was 0.89 (S1 Appendix, Table 2). All item loadings were significant, and latent factors were intercorrelated (Table 3).

Cluster correlations between valences

Correlational analyses were performed on the frequency sum scores of second-order clusters from PROC VARCLUS to examine relationships between unpleasant and pleasant clusters. Most correlations between unpleasant and pleasant clusters were not significant, and weak or close to 0 (Table 4). However, Parasympathetic Activation was significantly negatively correlated with 7 of 9 unpleasant clusters: Sympathetic Activation, Generalized Pain, GI Distress, Kinesthetic Disconnect, Low Energy, Neurological Discomfort, and Metabolic Stress (r = −0.26 - −0.40). In addition, there was a significant but weak inverse correlation between Activated & Attractive and Low Energy (r = −0.22) and a significant but weak positive correlation between Physical Intimacy and Heat & Urgency Response (r = 0.25).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Correlations between pleasant and unpleasant factors.

https://doi.org/10.1371/journal.pone.0353167.t004

Initial description of latent factors

Descriptive statistics of the final items in the unpleasant and pleasant latent factors can be found in Table 5. The exploratory t-test of global scores found global pleasant scores (M = 2.1, SD = 0.5) were higher than global unpleasant scores (M = 1.4, SD = 0.5), indicating pleasant items, on average, occurred more frequently than unpleasant items, t(273) = −18.3, p < 0.0001. All factors had complete cases from at least 75% of the sample.

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Descriptives of the final unpleasant and pleasant factors from CFA.

https://doi.org/10.1371/journal.pone.0353167.t005

Convergent validity and proof-of-concept

To assess convergent validity, the latent factors identified from CFA were associated with existing, conceptually relevant surveys using SEM. Figs 2 and 3 show all significant associations (see S1 Appendix, Tables 4 & 5 for full statistics). We hypothesized that unpleasant latent factors would be positively associated with premenstrual symptoms and negatively associated with body awareness; the opposite relationships were hypothesized for pleasant latent factors. As hypothesized, premenstrual symptom severity from Section 1 of the PSST was positively associated with all nine unpleasant latent factors (Fig 2) and negatively associated with three of the four pleasant latent factors (Physical Alignment, Physical Intimacy, and Parasympathetic Activation; Fig 3). In partial support for the hypotheses, BAQ scores were positively associated with Heat & Urgency Response (Fig 2) and all four pleasant factors (Fig 3).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Association of the unpleasant latent factors identified from the CFA (red circles) with scores from validated surveys and lifestyle behavior items (gray rectangles).

Each survey and lifestyle behavior was tested separately; significant associations from all models are condensed into a single figure to support proof-of-concept for factors. Solid lines denote positive associations; dashed lines denote negative associations. Patterned gray rectangles denote models tested only on a subset of participants (i.e., those who exercised or those who drank at all). Standardized estimates are presented for each significant association.

https://doi.org/10.1371/journal.pone.0353167.g002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Association of the pleasant latent factors identified from the CFA (blue circles) with scores from validated surveys and lifestyle behavior items (gray rectangles).

Each survey and lifestyle behavior was tested separately; significant associations from all models are condensed into a single figure to support proof of concept for factors. Solid lines denote positive relationships; dashed lines denote negative relationships. Patterned gray rectangles denote models tested only on a subset of participants (i.e., those who exercised or those who drank at all). Standardized estimates are presented for each significant association.

https://doi.org/10.1371/journal.pone.0353167.g003

As proof-of-concept, we expected pleasant latent factors to be positively associated with healthy behaviors (e.g., physically active, little-to-no alcohol consumption) and unpleasant latent factors to be positively associated with unhealthy behaviors (e.g., no exercise, moderate-to-high alcohol consumption). Consistent with these hypotheses, non-participation in exercise was positively associated with GI Distress and Kinesthetic Disconnect (Fig 2) and negatively associated with Physical Intimacy (Fig 3). Consuming any alcohol in the past 30 days was positively associated with Illness and Low Energy (Fig 2); interestingly, it was also positively associated with Physical Intimacy (Fig 3). A higher maximum number of drinks consumed in one drinking day (among drinkers only) was positively associated with Kinesthetic Disconnect and Metabolic Stress (Fig 2).

Participation in any resistance exercise was negatively associated with Sympathetic Activation and GI Distress (Fig 2) and positively associated with Physical Intimacy, Activated & Attractive, and Parasympathetic Activation (Fig 3). Greater resistance training exposure among resistance trained women was negatively associated with Sympathetic Activation, GI Distress, Kinesthetic Disconnect, Neurological Discomfort, and Heat & Urgency Response (Fig 2); however, it was also negatively associated with Physical Intimacy (Fig 3). Participation in aerobic exercise was positively associated with Generalized Pain (Fig 2); no other latent factors were significantly associated with aerobic exercise participation or exposure. Finally, relevant to the main goal of developing a woman-specific somatic inventory, we found that self-reported HC use was positively associated with Low Energy (Fig 2), Physical Intimacy, and Activated & Attractive (Fig 3).

Somatic experiences can be pleasant or unpleasant

Individuals typically seek treatment and healthcare to alleviate unpleasant sensations, which makes symptom-based assessments prominent in the existing literature. As expected, this made compilation of unpleasant items significantly easier than pleasant items. Expansion of the literature search to include items commonly used in mindfulness techniques, Eastern medicine (e.g., yoga), somatic therapy, and self-reported personal experiences was necessary to gather a sufficient number of pleasant items. Unsurprisingly, there were fewer pleasant items included in the models and fewer pleasant factors retained in the latent factors. Many of the pleasant items were related to physical appearance or more generalized and abstract (e.g., floating, activated, balanced) than unpleasant items, which often referenced specific, commonly experienced physical sensations (e.g., racing heart, headache, dizzy). Despite these challenges, the study achieved a primary goal of identifying four general pleasant domains that can be included in the subsequently developed inventory.

A key observation from assessing pleasant somatic sensations was that global pleasant frequency ratings were higher than global unpleasant frequency ratings. This suggests that pleasant sensations may be perceived as a steady, passive backdrop upon which salient, but perhaps more infrequent, unpleasant sensations punctuate lived experience and draw greater attention. This aligns with the Somatic Marker Hypothesis [1]; a static pleasant physical sensation may be a signal of a system/organism being at homeostasis or functioning at an optimal level. As a result, fewer cognitive and neural resources are allocated to these sensations. On the other hand, unpleasant sensations serve as signals of homeostatic deviation and, when of sufficient intensity, warrant the individual’s attention to correct or alleviate. The stronger a disruptive somatic sensation is, the more attention it draws, cuing decision-making and behavioral action to restore homeostasis [4].

The unpleasant factors were strongly related to one another, as were the pleasant factors. Further research is needed to understand whether such within-valence correlations reflect broader, generalized somatic states (i.e., when one system is out of homeostasis, the person feels “off”) or temperament differences (i.e., some people are oriented towards noticing negative versus positive states). Overall, these results parallel those observed for affect assessment (e.g., Positive and Negative Affect Schedule (PANAS)) [40]. Affect, similar to the somatic items within the present study, appears to consist of two unipolar dimensions, positive and negative, that are weakly related between valences but more strongly within [40,41]. In other words, the experience of unpleasant sensations can be fully distinct from the experience of pleasant sensations. Parasympathetic Activation, however, was the exception: it was inversely associated with seven of the nine unpleasant factors. Items in Parasympathetic Activation relate to a general state of homeostasis and wellness: feeling hydrated, relaxed, rested, comfortable, balanced, and healthy. When reported as occurring more frequently, these items appear to oppose unpleasant sensations. Taken together, these results point to the importance of conceptualizing bodily states as both pleasant and unpleasant, and somatic experiences as both tonic background and momentary experiences.

Somatic factors often capture integrated physiological states

GI Distress, Neurological Discomfort, and Kinesthetic Disconnect comprised items related to a single body system. Yet, several other unpleasant factors reflected typological clustering of somatic experiences rather than organ/body system-specific categories. Specifically, Sympathetic Activation, Metabolic Stress, Low Energy, and Heat & Urgency Response contained items that spanned more than one body system and instead related to an overall experience that is a composite of each item. To illustrate, Sympathetic Activation items span cardiac, respiratory, and sensory experiences that, together, reflect a common stress state – one that conceptually aligns with psychological states of anxiety and panic. Likewise, Metabolic Stress includes neurological, kinesthetic, and digestive symptoms, all of which parallel common hypoglycemic experiences. Finally, Low Energy items include kinesthetic, energetic, and digestive symptoms related to exhaustion. Thus, the somatic factors are a first step to tapping into higher order patterns of somatic experiences that might better align with both physical and mental health states. This has broad implications for research and clinical assessment of women’s health as it provides a new perspective on defining “well-being” both from an organ-specific, medicalized perspective and an interoceptive, cognitive perspective.

Generalized Pain and Illness appeared to reflect a general physiological domain, but with conceptually related items that may or may not occur concurrently. For example, the first-order cluster of muscle pain, joint pain, and feeling achy reflect musculoskeletal pain, which may or may not co-occur with the first-order cluster of abdominal cramps and lower abdominal pain that reflect digestive or menstrual discomfort. Replication with a larger, more heterogeneous sample (i.e., age range, demographics) is needed to determine whether a single Pain factor is optimal for describing how a woman experiences her body or whether multiple factors that differentiate pain subtypes should be explored. Likewise, Kinesthetic Disconnect items were broader in nature, and included symptoms related to proprioception (e.g., unstable), musculoskeletal agitation (e.g., restless, fidgety), and general physical discomfort (e.g., weak, uncomfortable) which may occur independently, or could, in fact, be related to a general state of unpleasant kinesthesia. Unfortunately, the retrospective method used in this study makes determination of within-factor item co-occurrence or independence impossible but suggests a need for future studies.

Similar observations about the typological clustering of items were made within pleasant factors. For example, whether feeling grounded means you also feel flexible (Physical Alignment); having full breasts means heightened sexual interest and enjoyment (Physical Intimacy); having a good hair day means you also feel radiant and resilient (Activated & Attractive); or being clean means you also feel rested and relaxed (Parasympathetic Activation) is unknown. Yet, it is possible that these somatic experiences share fundamental features that embody a particular state of being. Considerable work remains to further shape and hone the factors identified here and more fully capture how women experience their bodies. Still, the present study clearly illustrates the need to assess pleasant bodily sensations and experiences in women, especially related to healthy physiological sensations.

Somatic factors are conceptually linked to health and behavior

A primary goal of this study was to initiate the development of an instrument for eventual tracking of comprehensive somatic changes in women across reproductive life stages, in relation to health behaviors, and in response to mental and physical health interventions (e.g., tracking efficacy of substance use or exercise interventions). Thus, exploratory structural equation models were conducted to relate unpleasant and pleasant latent factors to women’s health, somatic, and behavior variables to provide proof-of-concept for the inventory.

The observed relationships of several unpleasant and pleasant latent factors with premenstrual symptoms provide initial validation for the preliminary survey items and the importance of considering somatic state in women’s health. Women who reported greater premenstrual symptoms also broadly reported more frequent unpleasant somatic experiences and less frequent pleasant sensations. This finding parallels prior findings that women with clinical-level menstrual symptoms report greater health anxiety and lower health-related quality of life [9] and may experience premenstrual exacerbation of existing physical and mental conditions (e.g., depression, anxiety, asthma, diabetes, migraines, and irritable bowel syndrome) [42,43]. Importantly, the present somatic items moved beyond classic negative menstrual symptom items to assess a wide range of body systems across a full spectrum of valence. This will allow greater precision in the assessment of individual differences in women’s somatic experiences, particularly in clinical samples where treatment may not only target a reduction in unpleasant sensations, but also an increase in pleasant sensations.

In a separate proof-of-concept analysis, pleasant and unpleasant latent factors were also associated with the use of hormonal contraceptives that included exogenous estradiol and/or progesterone. HC use was positively associated with Physical Intimacy, Activated & Attractive, and Low Energy, potentially implying more frequent experiences of these states among HC users. It is noteworthy that all participants in the HC and NC groups reported 25–35 day cycles; thus, this observation is not related to the absence of menstrual bleeding. Due to the nature of the study, it is not possible to determine whether these relationships were due to neuroendocrine effects of HC or differences in trait-like features related to HC use (e.g., sexual attitudes); however, women who used HC in this study were more likely to be White (versus Asian) and to have recently consumed alcohol compared to naturally cycling women, possibly alluding to mediating socio-cultural and/or behavioral effects. Further, HC use may make women feel safer engaging in sexual activity, thereby increasing the frequency of attracting and engaging in sexual activity [44]. Future research is needed to disentangle neuroendocrine effects from other effects. Moreover, additional research should investigate the utility of measuring somatic experiences in women of different ages and reproductive phases. For example, Heat & Urgency Response has the potential to detect hot flashes in perimenopausal women, with many other possible items of relevance to menarche and pregnancy.

As hypothesized, body awareness was positively associated with pleasant somatic states, supporting prior findings that body awareness positively contributes to physical health and well-being [16–18]. The items that comprise the Body Awareness Questionnaire (BAQ) gauge attentiveness to general physiological states and expectations from those states with vague valences that can be construed as either positively or negatively. Yet, all pleasant factors were associated with total scores from the BAQ. This further aligns our results with the Somatic Marker Theory, interoception, and allostasis theories, each of which predict that unpleasant sensations require less body awareness or interoceptive capacity to register in conscious experience compared to pleasant sensations [1,2,45,46]. The further development of the Women’s Somatic Experiences Inventory may also inform the current controversy regarding validity of many existing interoceptive measures [47–49], and embed women-specific topics in the nascent field of interoception research.

Exercise demonstrated any-versus-none and type-specific effects on somatic states. Not exercising had a dually negative effect. Conceptually, not exercising being positively associated with Kinesthetic Disconnect and negatively associated with Physical Alignment makes sense; those who are less active have less frequent encounters with experiences that require agility, flexibility, or strength (i.e., pleasant kinesthetic function) and may be more likely to experience kinesthetic discomfort. The findings that not exercising is positively associated with GI Distress and negatively associated with Physical Intimacy are relevant considering emerging evidence that sedentary behaviors can modify the gut microbiome and sexual functions [50,51]. Resistance, but not aerobic exercise, had a potent positive association with somatic sensations. Any amount of resistance exercise was positively associated with several pleasant and negatively associated with several unpleasant latent factors that covered a range of body systems, alluding to systemic benefits of participation in resistance exercise by young women. Further, among resistance exercisers, greater exposure yielded greater benefits, indicated by lower scores on five of the nine unpleasant factors. Research on the benefits of resistance training in women across the lifespan support widespread positive impacts on body composition, strength, muscle hypertrophy and preservation, cardiovascular health, and bone density [52,53]. These preliminary findings suggest that the further development of the Women’s Somatic Experience Inventory could benefit women’s health initiatives aimed at increasing physical activity and reducing sedentary behavior in women, especially as they age and become more susceptible to muscle and bone loss.

Finally, alcohol use had fewer associations with somatic factors than might be expected, possibly because the present sample was overtly healthy and mainly reported low-risk use patterns (i.e., 1 day per week, 2–4 drinks per drinking episode). The significant associations that were observed generally support prior studies that have linked immune function [54,55], sleep [56], and sexual activity [57] to alcohol consumption, but neither frequency nor quantity of consumption had any relation with the unpleasant or pleasant somatic factors. Interestingly, greater past 30-day maximal quantity in a single drinking episode was positively associated with Kinesthetic Disconnect and Metabolic Stress, but these factors included items that conceptually align with somatic experiences of intoxication and hangover: spinning, unstable, numb, shaky, weak, hunger pang.

Limitations

Because of the study’s exploratory nature, results should be interpreted in light of several limitations. Participants were college women, which limits generalizability to other samples of women, particularly considering college students often have a distinct set of health behaviors than women in other life stages. The present study was performed using the initial list of 300 somatic items and an iterative process that combined conceptual and theory-driven searches and quantitative clustering and path modeling techniques to create the preliminary factor structure of 73 items. The original item list of 300 was refined using pre-data and post-data conceptual steps, a data-driven step, and a post hoc conceptual step. Multiple imputation was performed with a Missing at Random assumption, but this was not explicitly tested. The CFI and TLI values for the CFA models for the unpleasant factors (and TLI for the pleasant factors) were below the range typically considered acceptable. Thus, results should be replicated in other samples and undergo additional rigorous psychometric and validity testing. However, considering the preliminary nature and iterative strategies used of inventory development, we propose that the pilot inventory (S2 Appendix File) may be a useful starting point for further study. For SEM models that evaluated associations between latent factors and within-behaviors (i.e., within drinkers or within exercisers), sample sizes were reduced and smaller than recommended, which may have inflated the estimates. Relevant confounding factors such as sexual activity, sexual orientation, religion, and other sociocultural factors were not assessed in this sample but may contribute to both the perception and interoceptive interpretation of somatic experiences. Moreover, comprehensive assessment of mental health concerns and unhealthy lifestyle behaviors were not fully assessed; these factors may affect the generalizability of the factors that were identified. Future research may wish to consider certain dropped somatic items that may be more pertinent to other life or reproductive stages. The names given to the identified factors may require additional refinement (e.g., Heat & Urgency Responses) and/or validation with physiological data (e.g., Parasympathetic Activation). Missing data were present across all factors due to participant’s ability to mark items as “Unclear” and non-requirement of full survey completion. Finally, the survey design consisted of a cross-sectional assessment of retrospective experiences; this introduced the potential for memory biases to skew the data. Nonetheless, this study provides a strong path forward to improve understanding of women’s subjective experiences of their bodies and uncover how these experiences relate to behavior and overall health.