International Journal of Behavioral Medicine

, Volume 19, Issue 1, pp 73–81

The Relationships Between Self-Rated Health and Serum Lipids Across Time

Authors

    • Faculty of ManagementTel Aviv University
  • Sharon Toker
    • Faculty of ManagementTel Aviv University
  • Samuel Melamed
    • Academic College of Tel Aviv-Yafo
  • Itzhak Shapira
    • Tel Aviv Sourasky Medical Center
Article

DOI: 10.1007/s12529-011-9144-y

Cite this article as:
Shirom, A., Toker, S., Melamed, S. et al. Int.J. Behav. Med. (2012) 19: 73. doi:10.1007/s12529-011-9144-y

Abstract

Background and Purpose

We studied the hypothesized effects of changes in self-rated health (SRH) on subsequently assessed changes in the levels of high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol, and triglycerides (TRI), separately for men and women. We also investigated the reverse causation hypothesis, expecting the initial changes in the levels of serum lipids to predict subsequently assessed changes in SRH levels.

Methods

We used a longitudinal design and controlled for possible confounders known to be precursors of both SRH and the above three serum lipids. Participants were apparently healthy men (N = 846) and women (N = 378) who underwent a routine health check at three points of time (T1, T2, and T3); T1 and T3 were on the average 40 and 44 months apart for the men and women, respectively.

Results and Conclusions

For the men, relative to T1 SRH, an increase in T2 SRH was associated with an increase in the T3 HDL-C levels relative to T2 HDL-C and with a decrease in the T3 TRI levels relative to T2 TRI. For the women, initial changes in the SRH levels did not predict follow-up changes in either of the lipids. For both genders, the reverse causation hypothesis, expecting the T1–T2 change in each of the serum lipids to predict T2–T3 change in SRH, was not supported. For the men, there is support for the hypothesis that the effects of SRH on morbidity and mortality, found by past meta-analytic studies, could be mediated by serum lipids.

Keywords

Self-rated healthHigh-density lipoprotein cholesterolLow-density lipoprotein cholesterolTriglyceridesLongitudinal design

Introduction

Several meta-analytic studies have concluded that self-rated health (SRH) predicts mortality and morbidity, including the incidence of cardiovascular disease, even after adjusting for traditional risk factors for cardiovascular disease, such as age and gender, and objective measures of health status, such as clinical diagnoses [3, 9, 21, 26]. Thus, accumulated evidence suggests that SRH reflects both obvious and subtle signs of disease progression, but the physiological mechanisms explaining the above predictive power of SRH remain poorly understood [23]. We argue that these signs of disease progression may include physiological processes affected by the level of serum lipids, such as atherosclerosis [12]. Following this rationale, in the current study we focus on three types of serum lipids: (a) high-density lipoprotein cholesterol (HDL-C), high values of which are considered to be protective against CVD [1]; (b) low-density lipoprotein cholesterol (LDL-C), a major risk factor in the etiology of cardiovascular disease (CVD; cf. [12]); and (c) triglycerides (TRI)—high values of which are causally implicated in the etiology of CVD [46, 48].

The physiological pathways linking SRH and subsequently assessed serum lipids probably involve the mediation of the sympathetic nervous system and the hypothalamic-pituitary-adrenal axis (HPA). Increased HPA activity following negative moods—such as worries that concern health—typically results in the secretion of catecholamines, cortisol, and glucagon, which in turn cause lipolysis and the subsequent release of fatty acids into the circulation [45]. For example, in a study of 50-year-old men in two countries, it was found that high levels of SRH were related to low concentrations of baseline (saliva) cortisol and to a strong cortisol response to stress [28]. An additional pathway leading from SRH to serum lipids could involve negative affect, specifically depressive symptoms. A recent study [27] found, based on a longitudinal design in which five waves of data were analyzed, that SRH unidirectionally predicted depressive symptoms while the reverse causation hypothesis was not supported. High levels of depressive symptoms were shown in a meta-analytic study [38] to be related to high levels of HDL-C and to low levels of LDL-C (the latter meta-correlation was not significant, however).

Considerable research attention has been directed toward understanding the information people incorporate into their self-ratings of health (e.g., [4, 5]). Following this research strand, a plausible explanation of the expected prediction of lipids by SRH is that people, when asked for a global evaluation of their health, take into consideration their own health behaviors, including dietary behavior that could lead to subsequent lipid elevations [47].

The associations between SRH and lipids have hardly been investigated. All past studies that we were able to identify used a cross-sectional design. With one exception [25] which found that HDL-C and LDL-C did not differ within SRH categories, these studies found linkages that are consist with our expectation. In a study of 4,005 men and women aged 71 or older conducted in the USA [24], HDL-C was found to be positively associated with SRH, after controlling for age, gender, and chronic diseases. In a study of 928 respondents aged 54 years and older in Taiwan [19], the ratio of total cholesterol to HDL-C among men was found to be associated with SRH. Following the above evidence, we hypothesized that:
  1. Hypothesis 1.

    Initial increases in SRH levels will predict subsequently assessed elevations of HDL-C and decreases of follow-up LDL-C and TRI levels.

     
Cholesterol plays an integral part in the structure and functions of the cell membrane and may also affect neurotransmission in the central nervous system thereby becoming implicated in the pathophysiology of mood disorders and health worries [33]. As we argued above, advanced atherosclerosis—as indicated by elevated levels of serum lipids—could predict subsequent lower levels of SRH. Based on the above rationale, we expected changes in the initial level of serum lipids to predict subsequently assessed changes in SRH. The only study in which baseline lipid levels were used to predict subsequently assessed SRH [17] did not confirm this unidirectional influence; however, these authors dichotomized SRH and used logistic regression to predict SRH, as opposed to the analytic approach of the current study in which SRH is regarded as a continuous variable. Therefore, we formulated the following hypothesis:
  1. Hypothesis 2.

    Initial increases in HDL-C levels and initial decreases in LDL-C and TRI levels will predict subsequently assessed increases in SRH levels.

     

Gender differences in serum lipids and in the psychological variables linking gender with lipids have been well documented [31, 53]. Physiological mechanisms that explain gender differences in serum lipids have been found (cf. [22]). Prior studies have found that SRH is a better predictor of subsequent mortality among men than among women [3, 21] and that men evaluate their health differently than women [4]. Different factors affect SRH among men and women [50] and SRH is differentially associated with health-related outcomes among men and women [6]. Following the above rationale and accumulated evidence, we analyzed our data separately for males and females. We expected our hypotheses to receive less support for women than for men because women’s SRH judgments are based on a wider range of health-related and non-health-related factors than those of men. However, we did not formulate a specific hypothesis regarding gender differences in SRH-lipids associations.

Methods

Study Participants

Study participants (N = 2,447, 1,714 men and 733 women) were all attending the Center for Periodic Health Examinations at the Tel Aviv Sourasky Medical Center for three routine health examination (T1, T2, and T3) between January 2003 and January 2010. These periodic health examinations were provided to the study participants by their employers as a subsidized fringe benefit: thus, attrition between T1 and T2 and between T2 and T3 could be due to change of employer, residence, or work location, and therefore totally unrelated to their participation in the current study. At T1, they represented 92% of the Center’s examinees during this period, all voluntary participants of the study. Non-participants did not differ from participants on any of the sociodemographic or biomedical variables. We also tested for attrition bias from T1 to T3. As compared with the study’s participants, those examined at T1 who did not return for a follow-up examination after about 2 years (59%) were more likely to be males, near retirement age, to have self-reported a chronic disease at T1, and to have reported spending less time in regular exercise activity at T1. We controlled for these possible sources of attrition bias in our data analyses.

We excluded respondents who self-reported having been diagnosed to have had a CVD (including stroke), diabetes, mental crisis, or cancer at either T1, T2, or T3, those who reported coming to the examination with known body inflammation, as well as participants who reported been diagnosed to have hyperlipidemia or regularly taking medications to treat any of the above chronic disease, including antidepressants, statins, fibrates, or any other lipid-lowering drug, aspirin, steroids, and antibiotics (N = 694 or 28%, 505 men and 186 women). We excluded this group of participants because the disease or the medication could impact the levels of the study’s criteria [51]. Obviously, individuals faced with the information from a healthcare provider that their health was chronically impaired would form congruent beliefs about their health. We also excluded several respondents who were not actively employed (mostly retirees) at T1, T2, or T3 (N = 120 or 5%, 55 men and 65 females), because retirement is known to affect one’s SRH [16]. A further 450 participants (18%), 308 men and 142 women, were excluded from the analysis because of missing data for one of the study parameters. Thus, the final sample consisted of 1,224 apparently healthy employees (50% of the initial sample, 846 men and 378 women). On the average, respondents were about 47 years old (SD ∼ 9 years), with about 16 years of formal education. The T3 wave occurred on average 40 and 44 months after T1 for the men and women, respectively; the T2 wave occurred on the average 20 and 22 months after T1 for the men and women, respectively.

Measures

Lipids

All lipids were assessed following a 12-h fast. Total cholesterol was determined enzymatically (cholesterol esterase followed by cholesterol oxidase) on Bayer Advia 1650; CV% according to the manufacturer is 1.4%. Triglycerides were determined by the Fossati three-step enzymatic reaction on Bayer Advia 1650; CV% according to the manufacturer is 1.6%. HDL-C was determined directly by a method developed by Izawa et al. on Bayer Advia 1650; CV% according to the manufacturer is 1.0%. LDL-C was calculated based on the levels of total cholesterol, HDL-C, and the triglycerides.

Self-Rated Health

SRH was measured by a single item, asking respondents to assess their general health with response options of excellent (scale value = 5), very good, good, fair, or poor (scale value = 1). This measure was found in several meta-analytic studies to be a valid tool for identifying persons with the greatest health needs and to have predictive validity relative to the criteria of all-cause morbidity and mortality (e.g., [10]). Additional information on the reliability and validity of the SRH measure as used in the current study is available (cf. [39]).

Control Variables

We controlled for possible confounders known to be precursors of both SRH and the above three serum lipids (cf. [2, 29]). Following the above rationale, we controlled for depressive symptoms, measured using the Personal Health Questionnaire-9 (PHQ-9), the depression section of a patient-oriented self-administered instrument derived from the PRIME-MD [41, 42]. The PHQ-9 lists nine potential symptoms of depression (e.g., feeling down, depressed or hopeless, little interest or pleasure in doing things), and asks patients to rate the frequency of experiencing each symptom during the past 2 weeks on a scale ranging from “never” to “almost always”. A meta-analytic study [18] concluded that it is as valid and reliable as longer clinician-administered instruments in a range of settings, countries, and populations.

Smoking Index

For cigarette smokers only, reflected the number of cigarettes smoked on a daily basis. A Physical exercise index was constructed based on the self-reported number of weekly hours regularly spent in intensive sports activities that caused pulse acceleration and sweating. Body mass index (BMI; kg/m2) was measured by a nurse and used as a continuous variable. Age and Gender were stated by the subject in the questionnaire.

Following evidence that the predictive power of SRH for subsequent morbidity and mortality risk may vary as a function of socioeconomic status [14, 52] we used a measure of financial strain as a control variable because it represents a major aspect of socioeconomic status. Financial strain was assessed by a six-item measure developed by Pearlin et al. [34] and often used in the literature. The six items asked respondents to report how often during the past month they have engaged in activities reflecting financial difficulties, like borrowing money to pay bills. Anchors ranged from “to a very large extent” (scale value = 5) to “not at all” (scale value = 1). Given the close association between serum lipids and dietary behavior [44], we also controlled for the number of servings of fruit and vegetables regularly consumed by the respondents. For the female respondents, adopting hormonal replacement therapy was based on self-report and was coded as a dummy variable with yes = 1. For the female respondents, using contraceptive pills was coded as a dummy variable with yes = 1.

Procedure

The local medical center ethics committee approved the study. An interviewer explained the survey to each participant and asked for her or his voluntary participation. Each participant signed a written informed consent form, which included our pledge of confidentiality. As part of the periodic health examination at the medical center, all respondents underwent serum sampling (after an overnight fast), anthropometric measurements, a physical examination, urinalysis, stress ECG and spirometry, vision and hearing function tests, and in addition completed a questionnaire. Following each health examination, a physician provided oral feedback to each respondent; a written feedback report with all the results of the periodic health examination was mailed to each respondent about 2 weeks after the examination.

Statistical Analyses

All criteria and predictors were systematically examined to detect outliers or non-normal distributions (i.e., skewness >2.0 and kurtosis >5.0); none was detected. The control variables of age, BMI, smoking index, physical exercise index, fruit and vegetables consumption, financial strain and depressive symptoms were based on their T1 values. For each respondent, we also calculated and controlled for the precise T1–T3 time lag in days. Using ordinary least squares regressions (with SPSS as software), we tested the study’s major predictions first by a set of regressions of the T3 levels of each HDL-C, LDL-C, and TRI on the control variables, the lipid T2 levels, and T2 and T1 SRH levels. This analytic technique provides support to the possibility that SRH is causally related to the lipids because changes in SRH levels antecede in time changes in the levels of blood lipids. In this set of regressions, we entered as the first set of predictors the control variables. They were followed by the T2 levels of each criterion; therefore, all other predictors predicted the T2–T3 change in the criterion [49]. Thereafter, we introduced into the regressions the T2 and T1 levels of SRH. When we tested the reverse causation hypothesis, from T3 SRH to the lipids, we used the same set of control variables, followed by T2 SRH levels, and the last two predictors were T2 and T1 levels of each of the lipids.

Results

Table 1 provides the means and standard deviations of all the study’s variables for the men and women. For testing the statistical robustness of our decision to stratify our analyses by gender, we tested the null hypothesis that the observed covariance matrices of the variables in our study, as described in Table 1, were equal across the two gender groups. Using Box’s Test [30], we found that the null hypothesis was rejected (Box’s M = 807.3, F = 3.41; p < .001). Additionally, we found that the mean vectors of the male and female respondents were significantly different from each other (Wilk’s Lambda = 0.66, p < .001 and Hotelling’s trace = 0.52, p < .001). Thus, we included in Table 1 the results of the separate t tests of the significance of the difference of each pair of male–female means (Table 1, last column).
Table 1

Means and standard deviations of the study’s variables for men and women

Variable

Men (n = 846)

Women (n = 378)

B (SEB)

β

ΔR2

B (SEB)

β

ΔR2

Criterion, T3 HDL-C

Control variables

  

0.10*

  

0.05*

 Age

0.03 (0.02)

0.02

 

0.02 (0.05)

0.02

 

 BMI

−0.25* (0.07)

−0.08

 

−0.07 (0.11)

−0.02

 

 Smoking index

−0.49 (0.45)

−0.02

 

0.09 (0.99)

0.01

 

 Physical exercise index

0.15 (0.11)

0.02

 

0.27 (0.25)

0.02

 

 Financial strain

−0.77* (0.37)

−0.05

 

0.49 (0.70)

0.02

 

 Time lag, T1–T3

0.01* (0.001)

0.06

 

−0.01 (0.01)

0.01

 

T2 level of the lipid

  

0.56*

  

0.55*

 T2 HDL-C

0.87* (0.02)

0.80

 

0.83* (0.03)

0.78

 

Predictors

  

0.04*

  

0.01

 T2 SRH

(0.47)

0.07

 

−1.25 (1.09)

−0.03

 

 T1 SRH

−0.78 (0.45)

−0.04

 

1.62 (1.06)

0.04

 

Total R2

  

0.70*

  

0.61*

Criterion, T3 LDL-C

Control variables

  

0.02*

  

0.20*

 Age

0.02 (0.08)

0.01

 

0.39* (0.12)

0.12

 

 BMI

−0.20 (0.20)

−0.02

 

−0.16 (0.25)

−0.02

 

 Smoking index

2.23 (1.34)

0.04

 

−0.56 (2.16)

−0.01

 

 Physical exercise index

−0.39 (0.34)

−0.03

 

−0.19 (0.57)

−0.01

 

 Financial strain

0.04 (1.20)

0.01

 

0.37 (1.15)

0.01

 

 Time lag, T1–T3

0.01* (0.003)

0.05

 

0.01* (0.003)

0.08

 

T2 level of the lipid

  

0.52*

  

0.37*

 T2 LDL-C

0.73* (0.02)

0.73

 

0.74* (0.04)

0.70

 

Predictors

  

0.01

  

0.01

 T2 SRH

−0.06 (1.43)

−0.01

 

−2.14 (2.36)

−0.04

 

 T1 SRH

−1.15 (1.37)

−0.03

 

−2.28 (2.32)

−0.03

 

Total R2

  

0.55*

  

0.58*

Criterion, T3 triglycerides

Control variables

  

0.06*

  

0.06*

 Age

0.23 (0.19)

0.03

 

0.04 (0.20)

0.04

 

 BMI

(0.50)

0.08

 

0.90* (0.43)

0.09

 

 Smoking index

−4.80* (3.32)

−0.04

 

3.59 (3.87)

0.04

 

 Phys. exercise index

−0.20 (0.86)

−0.01

 

0.21 (1.02)

0.01

 

 Financial strain

1.25 (3.00)

0.01

 

1.39 (2.72)

0.02

 

 Time lag, T1–T3

0.01 (0.01)

0.02

 

0.01 (0.01)

0.02

 

T2 Level of the lipid

  

0.31*

  

0.30*

 T2 TRI

0.56* (0.03)

0.59

 

0.61* (0.05)

0.58

 

Predictors

  

0.02*

  

0.01

 T2 SRH

−6.91* (3.25)

−0.06

 

−2.92 (4.24)

0.01

 

 T1 SRH

3.17 (3.39)

0.03

 

−1.62 (4.17)

−0.02

 

Total R2

  

0.39*

  

0.37*

Based on list-wise deletion of missing cases N’s are 845 and 378 for men and women, respectively

T3 time 3, T2 time 2, T1 time 1, HDL-C high-density lipoprotein cholesterol, LDL-C low-density lipoprotein cholesterol, TRI triglycerides, SRH self-rated health, BMI body mass index

Overall, the mean levels of lipids reported in Table 1 are comparable to those reported for other large samples in Israel [48]. For both genders, we found that while the mean vectors of HDL-C were not significantly different among T1, T2, and T3, they were significantly different for both LDL-C and TRI (Wilks’ Lambda = 0.54 and .64, p < .001, respectively). The decline we found in the means of LDL-C and TRI from T1 to T3 is consistent with the declining levels of serum lipids found in a representative sample of US adults [7].

We also inspected the correlation matrices of all the study’s variables for the men and women (not included here because of the size of the matrices; available upon request from the first or second authors). For the men, T1, T2, and T3 SRH measures were found to be significantly associated with all three HDL-C measures (mean r = .17). For example, the correlations between T3 HDL-C and SRH at T1 and T2 were .19 and .13, respectively. In comparison, for the women the majority of these correlation coefficients were not statistically significant. For example, for the women the correlations between T3 HDL-C and SRH at T1 and T2 were .09 and .02, respectively. For both genders, the correlations of T1, T2, and T3 SRH with T1, T2 and T3 LDL-C were for the most part small and not significant. For the men, T1, T2, and T3 SRH measures were significantly correlated with all three TRI measures (mean r = −.11). For example, T3 TRI correlated with T1 SRH and with T2 SRH; r = −.08 and −.14, respectively. In comparison, for the women the majority of these correlations were not statistically significant. For both genders, T1, T2, and T3 SRH measures were found to be negatively correlated with T1, T2, and T3 BMI (mean r = −.15) and depressive symptoms measures (mean r = −.24).

As is evident from Table 2, there was considerable support for our first hypotheses but only for the men, not for the women. The change from T1 to T2 in SRH was found to be a significant predictor of the change from T2 to T3 in HDL-C and in TRI but only among the men. We did not include fruit and vegetable consumption in Table 2 because these control variables were found to be insignificant predictors across all our regressions. Also, the two control variables used in our regressions on the subsample of women, being on hormone replacement therapy and habitually taking contraceptive pills, were not included in Table 2 because they were insignificant predictors in all our regression runs. Thus, for the male respondents, there is partial support for hypothesis 1. Following a reviewer’s suggestion, we examined the stability of the above results using as a control variable T1 levels of each lipid instead of its T2 levels. We obtained the same set of results reported in Table 2, with only slight changes, thus providing support for the analytic strategy used.
Table 2

The change from T1 to T2 SRH and control variables as predictors of the change from T2 to T3 in lipids levels

 

Men (n = 846)

Women (n = 378)

Variable

B (SEB)

β

ΔR2

B (SEB)

β

ΔR2

Criterion, T3 HDL-C

Control variables

  

0.10*

  

0.05*

 Age

0.03 (0.02)

0.02

 

0.02 (0.05)

0.02

 

 BMI

−0.25* (0.07)

−0.08

 

−0.07 (0.11)

−0.02

 

 Smoking index

−0.49 (0.45)

−0.02

 

0.09 (0.99)

0.01

 

 Physical exercise index

0.15 (0.11)

0.02

 

0.27 (0.25)

0.02

 

 Financial strain

−0.77* (0.37)

−0.05

 

0.49 (0.70)

0.02

 

 Time lag, T1–T3

0.01* (0.001)

0.06

 

−0.01 (0.01)

0.01

 

T2 level of the lipid

  

0.56*

  

0.55*

 T2 HDL-C

0.87* (0.02)

0.80

 

0.83* (0.03)

0.78

 

Predictors

  

0.04*

  

0.01

 T2 SRH

(0.47)

0.07

 

−1.25 (1.09)

−0.03

 

 T1 SRH

−0.78 (0.45)

−0.04

 

1.62 (1.06)

0.04

 

Total R2

  

0.70*

  

0.61*

Criterion, T3 LDL-C

Control variables

  

0.02*

  

0.20*

 Age

0.02 (0.08)

0.01

 

0.39* (0.12)

0.12

 

 BMI

−0.20 (0.20)

−0.02

 

−0.16 (0.25)

−0.02

 

 Smoking index

2.23 (1.34)

0.04

 

−0.56 (2.16)

−0.01

 

 Physical exercise index

−0.39 (0.34)

−0.03

 

−0.19 (0.57)

−0.01

 

 Financial strain

0.04 (1.20)

0.01

 

0.37 (1.15)

0.01

 

 Time lag, T1–T3

0.01* (0.003)

0.05

 

0.01* (0.003)

0.08

 

T2 level of the lipid

  

0.52*

  

0.37*

 T2 LDL-C

0.73* (0.02)

0.73

 

0.74* (0.04)

0.70

 

Predictors

  

0.01

  

0.01

 T2 SRH

−0.06 (1.43)

−0.01

 

−2.14 (2.36)

−0.04

 

 T1 SRH

−1.15 (1.37)

−0.03

 

−2.28 (2.32)

−0.03

 

Total R2

  

0.55*

  

0.58*

Criterion, T3 triglycerides

Control variables

  

0.06*

  

0.06*

 Age

0.23 (0.19)

0.03

 

0.04 (0.20)

0.04

 

 BMI

(0.50)

0.08

 

0.90* (0.43)

0.09

 

 Smoking index

−4.80* (3.32)

−0.04

 

3.59 (3.87)

0.04

 

 Phys. exercise index

−0.20 (0.86)

−0.01

 

0.21 (1.02)

0.01

 

 Financial strain

1.25 (3.00)

0.01

 

1.39 (2.72)

0.02

 

 Time lag, T1–T3

0.01 (0.01)

0.02

 

0.01 (0.01)

0.02

 

T2 Level of the lipid

  

0.31*

  

0.30*

 T2 TRI

0.56* (0.03)

0.59

 

0.61* (0.05)

0.58

 

Predictors

  

0.02*

  

0.01

 T2 SRH

−6.91* (3.25)

−0.06

 

−2.92 (4.24)

0.01

 

 T1 SRH

3.17 (3.39)

0.03

 

−1.62 (4.17)

−0.02

 

Total R2

  

0.39*

  

0.37*

The symbols B and β represent the unstandardized and standardized partial regression coefficients, respectively, and SEB stands for the standard errors of the former. ΔR2 (adjusted for degrees of freedom) stands for the incremental R2 for each of the three steps used in the regression analysis: control variables, T2 level of the lipid, and predictors. See explanatory note to Table 1 for the variables’ abbreviations used. The results presented are those of the last step of the OLS regression for each of the runs

*p < .05

We also tested the second hypothesis, expecting a reverse causation. For both genders, we did not find any support for this hypothesis (a table with the full results is available from the first or second authors upon request). Therefore, our second hypotheses did not receive support for either the men or the women in our sample.

Discussion

As expected by our first hypothesis, we found that for the men, when T2 SRH levels increased relative to T1 SRH levels, this change predicted an increase in the T3 HDL-C levels (controlling for T2 HDL-C) and with a decrease in the T3 TRI levels (controlling for T2 TRI). For the women in our sample, we did not find any support for our first hypothesis. Similarly, for both genders, there was no support for our second hypothesis, expecting that the change from baseline to T2 in either of the lipids would predict the subsequently assessed T2–T3 change in SRH.

SRH is often used as a criterion variable in clinical trials and as a key variable in national surveys assessing health indicators in several countries (cf. [26]). Therefore, higher levels of SRH could be regarded as an important goal. It is too early to suggest that interventions designed to increase SRH may end up influencing people’s levels of serum lipids because our study did not identify the mechanisms that may account for the linkages found across time for the men. We recommend that future research explores possible mechanisms not included in our study such as health behaviors, positive and negative affects, functional status [32], and physiological mediators described above. There was no support in our study for the expectation that interventions designed to lower serum lipids below a certain level (e.g., [20]) could lead to heightened levels of SRH. We argue that our findings that changes in SRH predict subsequently assessed levels of two serum lipids, albeit only for the men in our sample, contribute to our understanding of the linkages between SRH and physical health.

Our failure to find support for the first hypothesis among the women is consistent with a recurrent finding in the literature—that the association between SRH and mortality are stronger among men than in women [8, 14, 36]. As found by Deeg and Kriegsman [8], women’s responses to the item gauging SRH tend to be linked with health conditions that are disabling, rather than fatal, whereas men’s self-ratings tend to take the mortality risks and lifestyle factors into account in their responses to the same item. It is also possible that our finding reflect gender differences in sensitivity to symptoms and signs of disease and impairments [3].

The lack of any support for our second hypothesis, expecting the initial change in the levels of each of the lipids to predict subsequently assessed changes in SRH levels, is consistent with the findings of the only study which examined this hypothesis using a longitudinal design [17]. All other longitudinal studies of SRH and physiological risk factors focused on the paradigm of SRH as predicting mortality [23], sought to explain it by physiological risk factors, and therefore did not investigate the possibility of reverse causation. A possible explanation of the lack of support for our second hypothesis may be derived from the commonsense model of illness cognition (cf. [32]) which postulates that people are problem solvers who actively assess somatic changes and sensations and then form hypotheses about their potential meaning. The commonsense “self-diagnosis” an individual assigns to his or her somatic changes reflects the match between active appraisal process and the underlying schemata of various disease states. It is plausible that our respondents, in forming mental representations of their personal health, did not give appropriate weight to their knowledge of previous changes in their lipid levels or that these changes were not incorporated in their prototypical model of coronary vascular disease. Future research may investigate the impact of various types of credible information on changes in one’s lipid levels on subsequent changes in one’s SRH.

The current study has several strengths. Firstly, it is based on a large and heterogeneous sample. Secondly, we carefully excluded individuals who self-reported being chronically ill, including with hyperlipidemia, or taking medications that could influence their serum lipid levels. Very few past studies have applied these strict exclusion criteria. Thirdly, we tested the reverse causation hypotheses. Fourth, in our study, the time lag separating our T1 and T3 was, on the average, 40 months for the men and 44 months for the women. This time lag is well above the modal value used in past longitudinal studies relating psychosocial job characteristics to serum lipids [35, 40, 43]. However, we suggest that future researchers re-examine the possibility that initial changes in lipid levels may predict subsequently measured changes in SRH levels over longer periods of time. It could be argued that the effect sizes which we found for SRH, around 2% if assessed by the conservative criterion of ΔR2, is small. However, they represent distal effects found using a longitudinal design, and for people clinically considered “borderline” on their lipids level may mean being prescribed or not lipid-lowering medications. Additionally, by controlling for age, obesity, two types of health behaviors, depressive symptoms and above all the T2 levels of HDL-C and TRI, we further reduced our ability to explain a statistically significant share of the variances of lipids. For these reasons, we argue that the small effect sizes found represent one of the study’s strengths because they were based on a conservative test of the study’s hypotheses.

Yet study strength is our design which was based on three waves of repeated measures. This design has several obvious advantages. It controls for the confounding influence of time-invariant common causes [13]. It supports an interpretation of causality in the link between increased levels of SRH over time and the subsequent elevation or decrease in the levels of HDL-C and TRI among the men [54]. It enabled us to control for possible “third variables” like personality predispositions and genetic factors which influenced T1 measurements of SRH [37] and lipids [15]. Future research may submit this argument to an empirical test.

The results of this study should be interpreted with caution because of some limitations. Firstly, our sample of participants undergoing a periodic health examination may not be representative of the general population. Most of the individuals were highly educated white-collar workers who exhibited generally good health behavior patterns: they smoked little and exercised regularly. Owing to their superior health habits, our participants may have been more resilient to the effects of stress on SRH. However, it is even more likely that the significant findings obtained here with regard to SRH linkages with lipids will be replicated among less resilient respondents. Additionally, we controlled in our study for a major component of socioeconomic status, financial strain. Thirdly, as we noted above, several possible mechanisms may explain the effects of SRH on lipids. In the current study, we did not investigate the possible mediating mechanism. Testing such mechanisms would have helped to explain, on the physiological level, why for the men changes in SRH were found to be predictive of changes n HDL-C and TRI while the reverse causation hypothesis was disconfirmed for both genders. Testing these mechanisms appears to be a promising avenue for future research.

Yet another reservation concerns the finding, reported above, that on the average the mean LDL-C and TRI levels declined for both genders across the three waves of measurement. We noted that this decline reflect a general trend found in other advanced countries. An additional explanatory factor of the above finding could be that T1 participants changed their health habits between T1 and T3 due to the feedback they had received on their lipid levels following their T1 routine health checkup. In our analysis, we controlled for some but not all health habits: for example, we did not control for changes in diet. People attending periodic health examinations were shown to be inclined to change their health behaviors [11]. Therefore, we recommend that future research specifically cover possible changes in diet as related to changes over time in lipids levels.

In conclusion, our study adds longitudinal—time evidence to prior cross-sectional studies that found SRH to be associated with physiological risk factors (e.g., [19]). The effects of changes in SRH levels on subsequently assessed changes in HDL-C and TRI that we found, albeit only for the men, support the theoretical arguments that when formulating a subjective assessment of their health, people take into consideration information on the levels of objective risk factors. People’s subjective assessments of their state of health probably reflect subclinical physiological processes and also adverse health behaviors likely to impact their future lipid levels. Hence, the major task for future research is to explain the psychological and physiological mechanisms that mediate the across-time linkages found in our study.

Copyright information

© International Society of Behavioral Medicine 2011