Psychometric Analysis of the Health Self-Care Neglect Scale

Abstract

Objectives:

The Health Self-Care Neglect (HSCN) scale is a measure of self-care neglect developed for use in informal caregivers, where self-care is defined as behaviors undertaken to maintain health. There was no formal psychometric analysis of the scale, so we tested a 9-item, dichotomous response version of the HSCN scale in a sample of 250 informal caregivers of adults with chronic heart failure.

Methods:

As the indicators of self-care neglect were considered formative (influencing the latent variable directly) rather than reflective (influenced by the latent variable), we used a procedure for the specification of formative measurement models. First, maximally correlated composites of indicators were identified for the latent variable and optimal scoring weights were developed. Then, the reflective factor was tested with confirmatory factor analysis and longitudinal invariance of the factorial structure was tested by introducing model constraints. Reliability was assessed with composite reliability model-based estimates. Concurrent validity was assessed by correlating the HSCN scale total score with the maintenance scale score of the Self-Care Inventory.

Results:

Strict invariance, the highest level possible, was achieved. Reliability was .81 at baseline. Concurrent validity was demonstrated (r= − 0.475, p<0.0001).

Conclusion:

The results of this analysis indicate that the HSCN scale is reliable, stable, and valid as a measure of health self-care neglect when tested in a sample of caregivers. The HSCN scale measures the successful performance of self-care while existing measures of self-care reflect intention. Understanding both intention and behavior is useful, so we recommend using the HSCN scale in addition to existing measures of self-care.

Introduction

Cardiovascular disease (CVD) remains the primary cause of death and disability worldwide.¹ Although genetics are extremely important in causing CVD, unhealthy lifestyle behaviors are an independent contributor to the development of and poor outcomes associated with CVD.² For this reason, clinicians and researchers have focused on helping people to improve their self-care behavior.

Self-care behaviors include health promoting and illness management practices (see Box).³ Changing these behaviors is extremely challenging. Much is known about how to change self-care behaviors and a large body of research illustrates that much has been achieved over the years.⁴ However, many interventions such as motivational interviewing address the intention to change behavior rather than the behavior itself. Intentions address a behavioral goal and level of commitment to achieving that goal.⁵ Although self-care behavior and intentions are related, clearly people do not always do the things that they intend to do, for a variety of reasons. That is, people often voice their intention to behave in a particular way, but many do not follow through with that intention. This mismatch between intention and behavior is referred to as the intention-behavior gap.⁶

BOX

Terms Used

• Self-care behavior - behaviors undertaken to maintain health, consistent with the theoretical definition of self-care maintenance as health promoting and illness management practices.

• Intentions address a behavioral goal and level of commitment to achieving that goal.

• Self-care neglect – failure to engage in health promoting behaviors.

Ajzen captured the idea of an intention-behavior gap in the Theory of Planned Behavior, where he proposed that an individual’s decision to engage in a specific behavior can be predicted by their intention to engage in the behavior.⁷ Numerous studies have demonstrated that intentions predict behavior, but little of this work explicitly tests the relationship between intention and self-care behavior.⁵ This may be because most existing measures of self-care focus on individual intentions, asking respondents how likely they are to perform specific self-care behaviors (e.g., when you have symptoms, how likely are you to limit your activity until you feel better).^8–16 As intention is only one piece of the puzzle, a measure of specific self-care behaviors performed is needed to understand the intention-behavior gap.

In 2005 Boise and colleagues wrote about the Health Self-Care Neglect (HSCN) scale as a measure of neglecting to perform basic self-care behaviors.¹⁷ The original 10-item HSCN scale was adapted by Boise and colleagues from a health behaviors scale and used with informal caregivers to self-report the presence or absence of important health-related behaviors defined as behaviors undertaken to maintain health (i.e., going to the doctor, staying in bed when ill).¹⁷ The original scale, tested using data collected in 2001, had a reliability coefficient of .758.¹⁷ No other psychometric testing of the HSCN scale was located. Although promising, the scale was not used widely^18–20 and no formal psychometric analysis has been reported.

Self-care and self-care neglect are not simply the antithesis of each other. Engaging in self-care is an intentional act that may or may not be deliberative.^3,21 One’s own health is prioritized through behavior. Performing self-care sufficient to promote and maintain health improves outcomes across a range of chronic illnesses.⁴ Neglecting self-care may be intentional, caused by poor reasoning or insufficient reflection,³ or as in the case of caregivers, an inadvertent consequence of deprioritizing their own health while caregiving.²² Nearly a quarter of caregivers report difficulty taking care of their own health needs, with this percentage increasing for those under higher physical, emotional or financial strain.²² While there are many positive aspects of caregiving, measures quantifying self-care neglect are needed to better understand the health effects of caregiving.²³

The original HSCN scale had 10 items,¹⁷ but subsequent scale users measured only nine items, labeled as health-risk behaviors²⁰ or negative self-care behaviors,¹⁹ and discussed these behaviors as self-care. We used the modified 9-item version of the HSCN scale in an ongoing randomized clinical trial of a synchronous virtual health coaching intervention intended to improve the self-care of informal caregivers of adults with heart failure (R01 NR-018196).²⁴ The aim of the current study was to test the psychometric properties of the HSCN scale.

Methods

Sample

This study was a secondary analysis of existing data gathered from a consecutive sample of 250 informal caregivers of adults with heart failure (HF). These caregivers were recruited to participate in a longitudinal randomized controlled trial (RCT) assessing the efficacy of a support intervention designed to improve self-care.²⁴ Those receiving the intervention were compared to a control group receiving only vetted health information websites on heart failure and caregiving. The caregivers included in the RCT had to be providing informal care at least 8 hours/week, report high self-care neglect (HSCN scale score ≥ 2), and be able to complete the study protocol (e.g., adequate vision and hearing, willing to use technology, English speaking, no untreated major psychiatric illness). Caregivers were excluded from the trial if they demonstrated cognitive impairment (Telephone Interview for Cognitive Status²⁵ [TICS] < 25) or were participating in another support RCT.

Procedure

Caregivers were recruited primarily from a specialty HF clinic associated with a regional healthcare setting in a large urban city in the northeastern United States, supplemented with additional in-patient and other heart failure clinic sites in the same health system. Screening and enrollment were performed by research staff hired and trained for the study. Fully eligible caregivers provided informed consent and then completed the baseline assessment, either in-person or online prior to randomization. Follow-up data for the current study were obtained during phone calls performed at 1-month. The 1-month follow-up data were used to test the analytic solution for replicability.

Measurement

Two instruments were used to measure self-care in the RCT, the HSCN scale¹⁷ and the Self-Care Inventory;¹¹ both were used in this secondary analysis. The HSCN scale is a 9-item self-report instrument consisting of items measuring specific health maintenance behaviors. Respondents were asked if, in relation to their own health, over the prior 3 months they had neglected self-care (e.g., putting off going to the doctor, failing to get enough exercise, and taking medications improperly). The HSCN scale uses a dichotomous response format coded 0 for no and 1 for yes with individual items summed to yield a total scale score.¹⁷ Higher scores indicate more self-care neglect.

Additionally, we measured self-care using the Self-Care Inventory (SCI)¹¹ and used it for concurrent validity testing. The SCI, based on the Middle Range Theory of Self-Care of Chronic Illness,³ measures self-care maintenance, self-care monitoring, and self-care management. Only the self-care maintenance scale was used in validity testing because the 8-item SCI maintenance scale asks about healthy behaviors commonly used to maintain health (e.g., eat a balanced and varied diet, do something to relieve stress), similar to the behaviors measured in the HSCN scale. The SCI items are scored on a 5-point Likert-type scale ranging from never to always. Each item of the SCI maintenance scale is summed for a total scale score, which is then standardized to provide a score between 0 and 100. Higher scores indicate better self-care; 70 is considered adequate self-care maintenance.¹⁰

Analytic strategy

As the HSCN scale was not based on an explicit theoretical framework or even an explicit definition of the construct, the indicators of self-care neglect were considered formative (or causal) rather than reflective (or effect).²⁶ Both types of indicators are developed to correspond to a theoretical definition of a concept represented by a latent variable; however, they are distinguished by the direction of the influence linking the indicator to the latent variable(s). While formative indicators influence the latent variable directly, reflective indicators are influenced directly by the latent variable. While a set of reflective indicators should be correlated, formative indicators do not require such an association. This is a feature shared by many checklists, questionnaires, and measurement instruments used in health research.²⁷ A careful read of the HSCN scale items reveals that it is unnecessary for the items to be correlated; instead, it is plausible that the more of these behaviors that are manifested, the more evidence of self-care neglect can be seen.

Two issues raised regarding the implementation of measurement models based on formative indicators are particularly relevant: a) the absence of measurement error; b) the impossibility of statistical identification of the parameters of a formative measurement model when (as in our case) no paths are emitted from the latent variable. The second issue, in particular, renders it impossible to test formative measurement models per se within a traditional confirmatory factor analysis (CFA), unless they are embedded within a more complex nomological network where these latent variables exert their influence on other latent constructs or observed variables. These problems are overcome by the two-step approach devised by Treiblemeier, Bentler and colleagues.^28–30 This approach maintains the confirmatory perspective to measurement models and is consistent with the parcelling method described by Matsunaga³¹ as aggregating individual items into one or more “parcels”. Those parcel(s) are used as the indicator(s) of the target latent construct instead of individual items.

The two-step procedure of Treiblmaier and colleagues has two important advantages. First, it produces a measurement model that is statistically identifiable per se, without the need to be embedded in a wider nomological network where latent variables are influencing other latent variables. Second, it allows the inclusion of measurement error in the indicators of latent variables. In this approach, Step 1 is focused on the development of “item-parcels” to be used in Step 2 as indicators of latent variables. Parcels consist of maximally correlated composites of indicators identified by means of a canonical correlation analysis. In our case, the nine indicators were assigned to three canonical variates and optimal scoring weights for items were developed to obtain maximally correlated composites or parcels. Then, in Step 2, these optimally weighted/maximally correlated parcels were used as indicators of the HSCN latent variable in CFA.

Since the indicators of the HSCN scale are dichotomous, we used a non-linear canonical correlation analysis on the nine items defining the HSCN factor. The nine items were sorted into three sets by splitting each maximally correlated triplet of items into separate sets and optimal weights (i.e., canonical coefficients) for items were identified. Items 1, 4, and 9 were in the first set, items 2, 5, and 6 were in the second set, and items 3, 7, and 8 were in the third set.

These coefficients were then used to weight observed variables, forming the three composites to derive three indicators for the factor. As noted above, in step 2 CFA was used to test the measurement model.²⁸ To maximize the replicability of the factorial solution, the solution was tested using data collected at enrollment and 1-month later. Finally, longitudinal invariance of the factorial structure was tested introducing model constraints.^32,33

Non-linear canonical correlation analyses were performed with SPSS 27.0 (IBM Corp., 2020). After confirming the assumption of MCAR (Missing Completely at Random)³² with Little’s test (χ2(13)= 6.945, p=.905), CFA was performed using Mplus 8.9.³⁴ Model fit was evaluated by: (i) the chi-square test; (ii) comparative fit index (CFI)³⁵ and the Tucker and Lewis Index (TLI);³⁶ (iii) root mean square error of approximation (RMSEA)³⁷ along with the test of close fit and 95% confidence interval; and (iv) standardized root mean squared residual (SRMR).³⁸ Following Little’s³⁴ recommendations, we considered good fit for RMSEA values as below .05 and CFI and TLI values above .95. Following Hu and Bentler’s⁴⁰ recommendations, we considered a good fit for SRMR as values below .08.

After CFA provided a baseline for longitudinal invariance testing, instrument stability was assessed to determine if the psychometric characteristic of the scale indicators were consistent over time, which would allow the construct to be considered comparable across different occasions.^32,33 Longitudinal invariance was examined by comparing a series of nested models where a sequence of constraints was introduced on the parameters of the baseline model.³³ Chi square difference test (Δχ2)³⁹ and the difference in the CFI (ΔCFI)⁴⁰ were used to assess the tenability of these constraints. These constraints imply different levels of measurement equivalence, classified as Configural, Metric, Scalar, Strict invariance, providing increasingly more stringent tests for equivalence.

Configural invariance corresponds to the baseline model (no constraints) and explores whether a construct has the same meaning and basic factorial structure at different times, just positing the same number of factors and the same pattern of fixed and free loadings. This basic type of invariance provides empirical evidence of the qualitative similarity of factor structure. Metric invariance provides a more stringent test of equivalence by constraining factor loadings to be equal across time points, thus defining a common metric: the scale intervals are equal across times. Scalar invariance indicates that both factor loadings and intercepts of indicators are equal across time points. If an item shows scalar invariance, then differences in that item among time points are due only to differences across time in the construct the item is measuring; there is no item-bias across time. The final strict invariance level requires that factor loadings, intercepts and residual variances are the same across time points: in this case items are perfectly equivalent also in terms of measurement error across time.

Participant characteristics and HSCN instrument responses were summarized using means and standard deviations as well as frequencies and percentages for continuous and categorical variables, respectively. Concurrent validity of the HSCN scale was assessed by computing the Pearson correlation coefficient for the association between the HSCN scale total score and the maintenance scale of the Self-Care Inventory. Descriptive statistics and assessment of concurrent validity were conducted using SAS version 9.4 for Windows.

Results

The sample of 250 caregivers used in this analysis was predominately female (85.2%) spouses (59.8%), White (62.2%), and married (71.9%) (Table 1). Their mean age was 55.3 (SD 13.6) years. Most were well-educated with at least some college education (79.1%). Many of the caregivers were employed full-time (43.8%) and comfortable financially, reporting “enough” or “more than enough” family income (82.3%). Median reported time as caregiver was 3.5 years. See Table 2 for the percentage of the sample endorsing each item in the HSCN scale at baseline. At enrollment, data on the HSCN scale were available for 250 caregivers and data on the SCI were available for 249 caregivers. At 1-month, 94% of caregivers provided follow-up data (221 of 250) on both measures.

Table 1.

Caregiver Demographic Characteristics Self-Reported at Enrollment (N=250)

Race	Black or African-American	74 (29.7%)
	White or Caucasian	155 (62.2%)
	Asian	4 ( 1.6%)
	Other	5 ( 2.0%)
	Multi-racial	11 ( 4.4%)
Ethnicity (N=248)	Not Hispanic or Latino	239 (95.6%)
	Hispanic or Latino	9 ( 3.6%)
Sex	Male	37 (14.8%)
	Female	213 (85.2%)
Marital Status	Single	39 (15.7%)
	Married or partnered	188 (75.5%)
	Divorced/Separated	16 ( 6.4%)
	Widowed	6 ( 2.4%)
Caregiver Age in years	Mean	55.3
	STD	13.64
	Minimum / Maximum	19 / 79
Employment (N=248)	Employed full-time	109 (43.8%)
	Employed part-time	36 (14.5%)
	Unemployed	104 (41.8%)
Education (N=249)	Grade school	8 ( 3.2%)
	High school graduate or GED	43 (17.3%)
	Trade school or some college	83 (33.3%)
	Bachelor’s degree or more	104 (45.8%)
Finances (N=243)	Comfortable, more than enough	78 (31.5%)
	Enough to make ends meet	126 (50.8%)
	Not have enough to make ends meet	37 (14.9%)
Relationship with Patient	Spouse or Partner	158 (63.4%)
	Son, Daughter, Son in-law, Daughter in-law	36 (14.5%)
	Parent or Grandparent	35 (14.1%)
	Other relative	12 ( 4.8%)
	Friend	4 ( 1.6%)
	Other	4 ( 1.6%)
Time as a caregiver (years) (N=249)	Median	3.5
	Minimum / Maximum	0 / 46
Hours number of hours spent caregiving each day (N=247)	Mean	8.1
	STD	7.47
	Minimum / Maximum	0 / 24
Self-Care Inventory, Maintenance scale (N=249)	Mean	68.5
	STD	15.88
	Minimum / Maximum	6.3 / 100
Health Self-Care Neglect Scale	Mean	4.9
	STD	2.09
	Minimum / Maximum	2 / 9

GED = general education development

Table 2.

The Percentage of the Sample Endorsing Each Item in the Health Self-Care Behavior Scale at Baseline.

Health Self-Care Behavior Scale Item	Percentage of Sample Endorsing the Item at Baseline
1. Put off going to the doctor	52.4%
2. Failed to stay in bed when ill	63.1%
3. Postponed getting regular checkups or exams	52.8%
4. Cancelled or missed medical appointments	44.2%
5. Failed to get enough rest	69.6%
6. Taken medications improperly (too little, too much, not at all)	17.1%
7. Failed to get enough exercise	69.8%
8. Eaten poorly	53.6%
9. Put off recreational activities you enjoy (e.g., socializing with friends, attending church, etc.)	72.3%

The conceptual model of the indicators of the HSCN scale (Figure 1) had an excellent fit, χ2[N= 254, df =8] = 6.569, p=.58; RMSEA = .00, p=.88, [95% confidence interval (CI) limits = 0; .064]; CFI = 1; TLI = 1; SRMR = .025. All factor loadings were significant, ranging from .65 to .83; the factor correlation was .57. No constraints on parameters were specified across time or within the same time point.

Figure 1.

Final model of the indicators of the Health Self-Care Neglect scale

The composites here are parcels of items, not different facets of HSCN. cg_1 is composed of an optimal weighted sum of the items 1, 4 and 5. cg_2 is composed of an optimal weighted sum of items 2, 6 and 8, and cg_3 of items 3, 7 and 9.

Table 3 shows the results of the invariance analysis. Since all chi-square difference tests among adjacent nested models were statistically non-significant, all the posited constraints were tenable across time, providing evidence of excellent equivalence of the measures across a 30-day interval. This equivalence demonstrates the acceptability of comparing parameters and scores across times (e.g., comparing the latent as well as the observed means of the total scores across time). Figure 1 presents the standardized estimates of model parameters obtained in the more restrictive strict invariance testing. We used this solution for obtaining reliability estimates of the HSCN scale at time 1 (enrollment) and time 2 (1-month). Specifically, we used model-based estimates such as the composite reliability or Omega coefficient⁴¹ and the Maximal Reliability coefficient.⁴² Both of these coefficients were .81 at time 1. At time 2, the coefficients were .72 and .73 respectively for the HSCN scale.

Table 3.

Results of the Longitudinal Factor Invariance Testing

	χ2(p)	df	RMSEA [90%CI (p)]	CFI	CFI	SRMR
1. Configural	6.57 (.58)	8	0 [0; .053 (.88)]	1	1	.025
2. Metric	7.12 (.71)	10	0 [0; .051 (.94)]	1	1	.030
3. Scalar	8.15 (.77)	12	0 [0; .044 (.97)]	1	1	.036
4. Strict	9.57 (.85)	15	0 [0; .034 (.99)]	1	1	.034

Note. All chi square differences between subsequent models were non-significant.

Key: χ² chi-square; df degrees of freedom; RMSEA root mean square error of approximation; CFI comparative fit index; SRMR standardized root mean squared residual.

When assessing concurrent validity, the HSCN scale total score exhibited a negative correlation of moderate strength when compared to the SCI maintenance scale score (r=−0.475, p<0.0001). That is, as self-care behaviors rose, self-care neglect decreased.

Discussion

In this study we assessed the psychometric properties of a measure of self-care neglect. The results of this study indicate that the HSCN scale is reliable, stable, and valid as a measure of self-care behavior when tested in a sample of adult caregivers of people living with chronic heart failure. Further, these findings provide a conceptual foundation for the scale and validation against another valid and reliable measure of self-care. This examination of the HSCN scale properties adds to the available self-care measures in three important ways.

First, while the HSCN scale has been used in a few studies,^18–20 the lack of focus on its psychometric properties may have limited its use in research. This study is the first to present data on the psychometric properties of the HSCN scale, filling both a gap in the literature and providing support for other investigators’ use of this tool.

Second, the psychometric method used provides strong evidence of reliability, concurrent validity, and measurement equivalence. We found that the instrument was reliable, a valid measure of self-care theoretically defined, and stable with equivalent results across a 1-month interval. This equivalence is strong evidence that when parameters and scores are compared over time, they will reveal actual change in self-reported self-care behavior rather than measurement error. With this evidence, we advocate that investigators measure both intention and behavior whenever possible. Numerous measures of self-care intentions are available for the general population, individuals with various chronic conditions, and the caregivers who support them. The measures on this website https://self-care-measures.com/ are based on the Middle Range Theory of Self-Care of Chronic Illness, with scales measuring self-care maintenance, self-care monitoring, and self-care management.³ The HSCN scale can be considered a measure of self-care maintenance. Future research is needed to develop comparable simple measures of self-care monitoring and management behaviors. Doing so will facilitate our understanding of the relationship between self-care intentions and behaviors.

Third, measuring both intention as well as success (or failure) in performance allows investigators to explore what is responsible for this neglected behavior. We found that the self-care behaviors neglected most often in this sample of caregivers were recreational activities, exercise, and rest. Caregiving responsibilities may have used all available discretionary time for recreation, rest, and exercise, or perhaps the behaviors were viewed as unnecessary. This question cannot be answered from this analysis. However, in the future, understanding the contributors of both intention and behavior would facilitate development of interventions that empower people to perform self-care. The theory of reciprocal determinism,⁴³ which argues that behavior, cognition, and environment all interact and influence each other, would suggest that the more people engage in self-care, the more they will believe that they are capable of performing self-care in the future.

We believe that the results of this study can be useful to practicing clinicians. We recommend that health care professionals use this brief tool as a screener to assess if someone is engaging in high levels of neglect, or as a well to track progress toward reducing self-care neglect.

Limitations

While an important and novel contribution to the literature, there are some limitations of this study. First, the study participants in this trial were enrolled from one geographic region of the United States (southeastern Pennsylvania, both urban and suburban) and a single academic healthcare system. Second, the sample was predominantly female (85%) and well-educated (79.1% with some college or graduate school), which is higher than recently reported U.S. data (female caregivers: 61%; education some college or more: 35%)²² and a recent systematic review of heart failure caregiver burden (among the U.S. studies 63% female caregivers).⁴⁴ The uniqueness of our sample may be due to the specific healthcare system or clinic settings where the parent study²⁴ recruited.

Conclusions

Study of the “intention-behavior gap” – the failure for intentions to translate into behaviors or actions⁶ – requires the use of reliable and valid tools to capture both self-care intention and self-care behavior. Examining both intention and behavior using valid and reliable measures can help investigators better understand the “intention-behavior gap” and tailor interventions to their target population. We have demonstrated that the HSCN scale is a valid and reliable measure of self-care maintenance behaviors. We recommend using both existing measures of self-care (e.g., SCI) and the HSCN scale together in future research.