Association of childhood loneliness with incident hypertension, heart disease, and stroke among low-educated Chinese women

Abstract

Early-life psychosocial factors may be associated with long-term cardiovascular health. We investigated whether self-reported loneliness during childhood is associated with an increased risk of developing hypertension, heart disease, and stroke in later life. We focused on low-educated, middle-aged, and older women, a potentially vulnerable population. This study was conducted and reported in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) statement for cohort studies. We conducted a prospective cohort study using data from the China Health and Retirement Longitudinal Study (CHARLS), including 4341 women aged ≥45 years with low educational attainment (primary school or less) at baseline (2011) and followed through 2020. Childhood loneliness was assessed retrospectively using 2 questions on the frequency of loneliness due to having no friends and the presence of a close friend during childhood. Participants reporting frequent loneliness and no close friend were classified as exposed. Incident hypertension, heart disease, and stroke were defined by new physician diagnoses or blood pressure measurements/medication use. Cox proportional hazards models estimated hazard ratios (HRs) adjusted for demographic, behavioral, and clinical covariates. Mean baseline age was 61.21 ± 9.63 years. Over a median 9.0-year follow-up, 2245 women developed hypertension (52.44%), 1343 developed heart disease (31.37%), and 474 had a stroke (10.92%). Childhood loneliness was reported by 5.25% of participants. In fully adjusted models, childhood loneliness was associated with a 31% higher hazard of hypertension (HR 1.31, 95% confidence interval [CI]: 1.01–1.70) and a 41% higher hazard of heart disease (HR 1.41, 95% CI: 1.06–1.89). The association with stroke was elevated but not statistically significant (HR 1.47, 95% CI: 0.97–2.21). Childhood loneliness is associated with increased risks of hypertension and heart disease in later life among Chinese women with low education. This highlights the importance of early psychosocial experiences for cardiovascular health.

1. Introduction

Weak social connections (e.g., social isolation or loneliness) are associated with poorer health and longevity, with risks comparable to traditional factors.^[1] Research shows that weak social ties increase mortality risk, with a review indicating a 50% higher likelihood of premature death.^[2] A review found loneliness and social isolation in adulthood are associated with a 29% increased risk of coronary heart disease and a 32% increased risk of stroke.^[3] These findings underscore the profound association between social and emotional well-being and cardiovascular health.

Life-course epidemiology indicates that early experiences shape adult health, with adverse childhood experiences (ACEs) linked to higher chronic disease risks such as heart disease and diabetes.^[4-6] Studies show that childhood psychosocial adversity is also associated with elevated cardiovascular morbidity and mortality in adulthood.^[7-9] In particular, childhood social isolation has been linked to poorer cardiovascular health in Chinese adults in a recent nationwide cohort study.^[10]

However, within the broader domain of childhood adversity, the specific role of childhood loneliness – a subjective feeling of social isolation or lack of companionship in childhood—remains understudied and poorly understood. Loneliness is conceptually distinct from objective social isolation; it refers to the distressing experience that arises when one’s social relationships are perceived as inadequate or lacking.^[11] While adult loneliness has been linked to worse cardiovascular and mental health outcomes (e.g., depression, hypertension, coronary disease, and stroke), there is a dearth of information on the health outcomes associated with loneliness at earlier life stages.^[10] Childhood and adolescence are critical periods when social and emotional development occurs, and loneliness in youth might be associated with “pre-disease pathways” and later health outcomes.^[12,13] One recent study in a Chinese population found that self-reported childhood loneliness was associated with faster cognitive decline and a higher risk of dementia in adulthood.^[14] That study also noted that the association of childhood loneliness persisted even among individuals who did not report feeling lonely in adulthood,^[14] suggesting that early loneliness has a lasting independent association. Given these insights, it is plausible that childhood loneliness is also associated with cardiovascular risk in later life, potentially through mechanisms such as chronic stress responses or the adoption of unhealthy behaviors.^[15]

To our knowledge, no prior longitudinal study has directly examined the relationship between loneliness in childhood and the incidence of cardiovascular diseases (CVD) in older adults. We aimed to address this gap by leveraging data from a large national cohort in China. Specifically, we investigated whether older women with low education who experienced loneliness in childhood have higher risks of developing hypertension, heart disease, and stroke, compared to those who did not experience childhood loneliness. We focused on low-educated women because this group may be particularly vulnerable – women are known to be more affected by early psychosocial adversity in terms of later CVD risk,^[16] and lower socioeconomic position could exacerbate the long-term associations of childhood stress.^[17] We hypothesize that childhood loneliness is associated with a higher incidence of hypertension, cardiovascular events, and stroke in this population, after adjusting for adult risk factors.

2. Materials and methods

2.1. Study design and population

Data for this study were derived from the China Health and Retirement Longitudinal Study (CHARLS), a nationally representative longitudinal cohort of middle-aged and older adults in China. We utilized data from the baseline survey in 2011 and follow-up waves through 2020 (specifically, waves in 2013, 2015, 2018, and 2020). CHARLS is a nationwide questionnaire survey conducted in 150 counties across 28 provinces (autonomous regions and municipalities) in China, established by the National Development Research Institute of Peking University.^[18] Briefly, participants were enrolled in the baseline survey using a probability proportional to size sampling strategy, with 17,708 individuals undergoing questionnaires and physical examinations. The CHARLS study was approved by the Biomedical Ethics Review Committee of Peking University (IRB00001052-11015), and all participants provided written informed consent.^[18] As the present analysis used only fully anonymized, publicly available secondary data from CHARLS, no additional ethical approval or informed consent was required. Data are publicly accessible from the CHARLS project website.

Of the 4341 retained participants, childhood loneliness data were fully ascertained (excluded earlier), baseline stroke status was fully ascertained for all participants, and baseline hypertension/heart-disease status was each available for 4281 (the remaining 60 had missing baseline status and therefore had no predefined “event-free” baseline for that specific outcome). Accordingly, the analytic sample comprised 4341 women for the stroke outcome, 4281 for new-onset hypertension, and 4281 for new-onset heart disease. The complete-case approach described in Section 2.5 was applied to covariates with low missingness, while missing-indicator categories were retained for body mass index (BMI) and annual household income to preserve sample size; analytic sample sizes therefore varied slightly across models. The participant flow chart is shown in Figure 1.

Figure 1.

Flowchart of participant selection. *Of 17,708 participants in the CHARLS 2011 baseline, 13,367 were sequentially excluded: aged < 45 years (n = 648), male (n = 8322), non-low education (n = 3583), baseline cardiovascular disease (hypertension, heart disease, or stroke) or missing baseline data (n = 60), missing childhood loneliness (n = 747), and missing new-onset outcome data or no follow-up by 2020 (n = 7). The final analytic cohort comprised 4341 individuals. Low education was defined as primary school or less, consistent with prior CHARLS-based studies. CHARLS = China Health and Retirement Longitudinal Study.

2.2. Assessment of childhood loneliness

The primary exposure, childhood loneliness, was assessed retrospectively at baseline (2011) through self-report. CHARLS included 2 key questions about participants’ social experience in childhood^[14]: 1. “When you were a child, how often did you feel lonely because you had no friends? Was it often, sometimes, not very often, or never?” 2. “When you were a child, did you have a close friend (yes or no)?” Based on responses to these items, we classified participants according to the definition used in a recent publication on CHARLS.^[14] Childhood loneliness was defined as reporting both frequent feelings of loneliness (“often”) and the absence of a close friend during childhood.^[14] Participants who met only one of these criteria (e.g., answered that they often felt lonely but did have at least 1 close friend, or vice versa) were classified as having possible childhood loneliness. Thus, the exposure was classified into 3 mutually exclusive groups: no childhood loneliness (met neither criterion), possible childhood loneliness (met 1 criterion), and childhood loneliness (met both criteria). In primary analyses, the 3-level exposure was entered into Cox models with no childhood loneliness as the reference group; hazard ratios were estimated separately for the possible and childhood loneliness groups (Tables 2–5 and Fig. 2). An alternative binary classification (meeting both criteria vs meeting neither or 1 criterion) was evaluated as a sensitivity analysis (Table S2, Supplemental Digital Content 1). This strict definition captures individuals who experienced pronounced and frequent loneliness in their early life. In the overall CHARLS cohort, this definition corresponds to a small proportion – about 4.2% of participants were classified as having childhood loneliness in a prior analysis.^[10] In our low-educated female sample, the prevalence was slightly higher (as many older Chinese women had limited schooling and childhood in rural settings, factors associated with loneliness),^[10] but still only a minority (estimated under 10%) of the women were categorized as having childhood loneliness at baseline.

Table 2.

Association of childhood loneliness with new-onset heart disease, hypertension and stroke.

Characteristic	Model 1		Model 2		Model 3
	HR (95% CI)	P-value	HR (95% CI)	P-value	HR (95% CI)	P-value
New-onset heart disease
No childhood loneliness	Reference		Reference		Reference
Possible childhood loneliness	1.20 (1.03–1.38)	.017	1.18 (1.02–1.37)	.028	1.19 (1.02–1.38)	.022
Childhood loneliness	1.47 (1.10–1.96)	.009	1.40 (1.05–1.87)	.023	1.41 (1.06–1.89)	.020
New-onset hypertension
No childhood loneliness	Reference		Reference		Reference
Possible childhood loneliness	0.99 (0.87–1.13)	.900	0.95 (0.84–1.08)	.460	0.96 (0.85–1.09)	.546
Childhood loneliness	1.33 (1.02–1.72)	.033	1.27 (0.98–1.65)	.069	1.31 (1.01–1.70)	.042
New-onset stroke
No childhood loneliness	Reference		Reference		Reference
Possible childhood loneliness	1.23 (0.99–1.53)	.057	1.18 (0.95–1.46)	.140	1.16 (0.94–1.45)	.170
Childhood loneliness	1.55 (1.03–2.33)	.035	1.47 (0.97–2.21)	.066	1.47 (0.97–2.21)	.067

Model 1: unadjusted.

Model 2: adjusted for age, marital status, area of residence, drinking, smoking, annual income groups.

Model 3: adjusted for age, marital status, area of residence, drinking, smoking, annual income groups, BMI groups, disability.

Cox proportional hazards models were used to estimate HRs and 95% CIs.

BMI = body mass index, CI = confidence interval, HR = hazard ratio.

Table 5.

Stratified associations between childhood loneliness and new-onset stroke.

Subgroups	n	New-onset stroke, hazard ratio (95% CI) P-value					P for interaction
		No childhood loneliness	Possible childhood loneliness			Childhood loneliness
Age (yr)							.855
45–55	1230	Reference	1.30 (0.80–2.12) .293			1.96 (0.68–5.63) .210
56–63	1544	Reference	1.23 (0.86–1.74) .257			1.25 (0.63–2.49) .519
>63	1472	Reference	1.04 (0.74–1.46) .824			1.52 (0.84–2.74) .168
Marital status							.708
No spouse	809	Reference	1.26 (0.77–2.07) .364			1.08 (0.37–3.13) .888
Married with spouse	3437	Reference	1.11 (0.87–1.42) .395			1.48 (0.95–2.32) .082
Area of residence							.899
Urban	1147	Reference	1.07 (0.71–1.62) .736			1.42 (0.60–3.37) .428
Rural	3099	Reference	1.20 (0.93–1.55) .168			1.45 (0.91–2.32) .118
Diabetes							.181
No	4005	Reference	1.10 (0.87–1.37) .430			1.49 (0.98–2.26) .063
Yes	241	Reference	1.86 (0.84–4.14) .126			0.70 (0.08–5.82) .742
Heart disease							.500
No	3722	Reference	1.25 (0.98–1.60) .074			1.55 (0.98–2.45) .059
Yes	524	Reference	0.94 (0.58–1.51) .790			1.24 (0.47–3.24) .662
Arthritis							.121
No	2506	Reference	0.96 (0.71–1.29) .785			1.20 (0.65–2.22) .553
Yes	1740	Reference	1.49 (1.08–2.06) .015			1.85 (1.06–3.25) .032
Dyslipidemia							.524
No	3797	Reference	1.18 (0.93–1.50) .168			1.49 (0.95–2.32) .079
Yes	344	Reference	1.23 (0.68–2.25) .495			1.28 (0.36–4.58) .709
Missing	105	Reference	0.15 (0.01–1.69) .126			1.93 (0.06–60.39) .708
Drinking							.426
Never	3502	Reference	1.23 (0.97–1.57) .086			1.56 (1.00–2.43) .051
Current	525	Reference	0.96 (0.51–1.81) .898			1.31 (0.37–4.65) .679
Ever	219	Reference	0.49 (0.19–1.25) .135			0.57 (0.06–5.07) .613
Smoking							.608
Never	3837	Reference	1.11 (0.88–1.40) .372			1.21 (0.76–1.93) .418
Current	297	Reference	1.57 (0.67–3.68) .299			3.44 (1.03–11.46) .044
Ever	112	Reference	1.27 (0.37–4.33) .705			2.62 (0.41–16.67) .309
Annual income							.338
<15,000	1808	Reference	1.12 (0.82–1.55) .475			1.03 (0.54–1.97) .923
15,000–25,000	341	Reference	1.79 (0.86–3.70) .117			2.07 (0.44–9.61) .355
≥25,000	739	Reference	0.82 (0.47–1.42) .481			1.54 (0.53–4.48) .430
Missing	1358	Reference	1.39 (0.91–2.12) .125			2.52 (1.25–5.07) .009
Depressive symptoms							.638
No	2036	Reference	1.12 (0.79–1.59) .517			1.56 (0.79–3.08) .199
Yes	1996	Reference	1.30 (0.97–1.75) .078			1.30 (0.76–2.22) .338
BMI							.931
Underweight < 18.5	316	Reference	1.25 (0.45–3.45) .67			0.88 (0.09–8.29) .912
Normal weight 18.5–24	1792	Reference	1.22 (0.86–1.75) .265			1.27 (0.62–2.60) .516
Overweight or obesity > 24	1492	Reference	1.10 (0.79–1.53) .586			1.37 (0.71–2.63) .348
Missing	646	Reference	1.04 (0.59–1.83) .902			2.08 (0.87–4.97) .101
Fasting blood glucose (mg/dL)							.083
<97.02	1035	Reference	1.11 (0.71–1.72) .6578			1.42 (0.65–3.12) .3782
97.02–108.37	1041	Reference	1.25 (0.80–1.94) .3213			0.42 (0.10–1.78) .2414
>108.37	1057	Reference	0.86 (0.57–1.29) .4598			1.62 (0.78–3.35) .1956
Total cholesterol (mg/dL)							.231
<180.55	1044	Reference	0.82 (0.51–1.32) .417			1.07 (0.41–2.78) .885
180.55–212.25	1047	Reference		1.27 (0.83–1.93) .271	0.61 (0.19–1.99) .411
>212.25	1048	Reference		1.06 (0.70–1.60) .776	1.81 (0.91–3.60) .091
Triglyceride (mg/dL)							.306
<89.39	1044	Reference		1.48 (0.88–2.49) .140	1.06 (0.31–3.55) .931
89.39–140.72	1047	Reference		1.06 (0.70–1.61) .776	0.86 (0.34–2.21) .757
>140.72	1049	Reference		0.82 (0.56–1.20) .299	1.32 (0.67–2.57) .424
HDL-C (mg/dL)							.663
<44.45	1027	Reference		0.87 (0.58–1.30) .486	0.84 (0.33–2.14) .711
44.45–56.45	1064	Reference		1.23 (0.80–1.88) .347	1.54 (0.70–3.40) .285
>56.45	1052	Reference		1.15 (0.72–1.81) .561	1.48 (0.61–3.61) 0.390
LDL-C (mg/dL)							.445
<103.61	1034	Reference		0.91 (0.58–1.44) .696	1.11 (0.46–2.69) .821
103.61–132.65	1056	Reference		0.91 (0.59–1.40) .668	0.66 (0.23–1.86) .433
>132.65	1046	Reference		1.34 (0.88–2.04) .166	1.70 (0.80–3.59) .164
C-reactive protein (mg/dL)							.244
<0.64	1030	Reference		0.80 (0.48–1.34) .401	1.15 (0.48–2.71) .756
0.64–1.59	1063	Reference		1.02 (0.68–1.55) .911	0.19 (0.03–1.38) .100
>1.59	1049	Reference		1.20 (0.81–1.77) .361	2.11 (1.07–4.16) .031
Disability							.440
No	3367	Reference		1.23 (0.95–1.58) .112	1.32 (0.80–2.20) .276
Yes	872	Reference		1.03 (0.66–1.60) .903	1.90 (0.93–3.89) .079

Above model adjusted for age, marital status, area of residence, drinking, smoking, annual income, BMI, disability. In each case, the model is not adjusted for the stratification variable.

BMI = body mass index, CI = confidence interval, HDL-C = high-density lipoprotein cholesterol, HR = hazard ratios, LDL-C = low-density lipoprotein cholesterol.

Figure 2.

Multivariable-adjusted cumulative hazard functions for the association of childhood loneliness with new-onset heart disease (A), hypertension (B), and stroke (C). The Y-axis shows cumulative hazard, estimated as 1 minus the Kaplan–Meier event-free survival estimate (1 − S(t)) within each childhood-loneliness category. The X-axis covers 0 to 115 months, reflecting the 2011 to 2020 follow-up window. P-values are from multivariable Cox proportional hazards models (model 3, Table 2) adjusted for age, marital status, residence, drinking, smoking, annual income, BMI, and disability (A: P = .020; B: P = .042; C: P = .067). Because the tail reflects small at-risk numbers, curves are most reliably interpreted between 0 and 105 months. BMI = body mass index.

2.3. Outcome assessment

We examined 3 incident cardiovascular outcomes during follow-up: hypertension, heart disease, and stroke. All participants were asked about medical history and underwent certain measurements at each survey wave. We defined each outcome as follows:

2.3.1. Hypertension

Participants were considered to have incident hypertension if they were free of hypertension at baseline (no self-reported physician diagnosis and blood pressure below hypertension thresholds) and subsequently either had a blood pressure measurement in the hypertensive range or received a new physician diagnosis of hypertension during follow-up. Blood pressure was measured by trained staff in CHARLS using a standard oscillometric device at each visit. We defined hypertension as systolic blood pressure ≥ 140 mm Hg or diastolic blood pressure ≥ 90 mm Hg, or current use of antihypertensive medications. Baseline hypertension status was determined accordingly, and new cases during follow-up were recorded at the visit when criteria were first met.

2.3.2. Heart disease

CHARLS asked participants whether a doctor had ever told them that they have “heart disease,” including conditions such as heart attack (myocardial infarction), coronary heart disease, angina, or other heart problems. For our analysis, participants were followed for new self-reported diagnoses of heart disease at follow-up waves. An incident case was defined by a new affirmative answer indicating a physician diagnosis of heart disease since the last interview. While self-reported, similar questions have been used in large-scale epidemiologic studies and have reasonable validity for major events.

2.3.3. Stroke

Stroke history was likewise self-reported based on a physician’s diagnosis. Participants were followed for new self-reported doctor-diagnosed stroke in subsequent surveys. Incident stroke was defined by a first-time report of stroke during follow-up.

Importantly, the reliability of self-reported physician-diagnosed hypertension and diabetes in CHARLS has been formally validated against biomedical measurements in earlier work, showing high specificity (>96%) and moderate agreement (κ = 0.57–0.65) with objective biomarkers for these 2 conditions.^[19] For the additional outcomes analyzed in our study (heart disease and stroke), no comparable comprehensive validation study in CHARLS has been published to date; non-differential outcome misclassification with respect to childhood loneliness would therefore bias the reported effect estimates toward the null. In addition, hypertension in our analysis was ascertained using standardized clinical blood pressure measurements and medication use, thereby strengthening the credibility of the observed association for this primary outcome.

2.4. Assessment of covariates

We included several baseline covariates as potential confounders or modifiers, based on prior literature on childhood adversity and health.^[6,20] These covariates (all measured at the 2011 baseline) were: age (years, continuous), marital status (married/cohabitating vs not), area of residence (rural vs urban), smoking status (current smoker vs nonsmoker), and alcohol drinking (current drinker vs nondrinker, where “current” was defined as drinking alcohol more than once a month). Additionally, annual household income (in quartiles, to represent socioeconomic position in adulthood), BMI was categorized as underweight (<18.5 kg/m²), normal weight (18.5–24.0), or overweight or obesity (≥24.0), according to Asian cut-points; a separate category for missing values was retained to preserve sample size for stratified analyses (Tables 3–5), and disability status were included. Disability was defined as having any impairment in basic activities of daily living (ADLs) or instrumental ADLs, or self-reported, doctor-diagnosed disability. These factors were selected to adjust for demographics and health behaviors that may confound or mediate the relationship between early loneliness and later cardiovascular health. Notably, education level was not included as a covariate because the analysis was restricted to a narrow education group (low education), thus maintaining relatively constant baseline education by design. Continuous biochemistry variables (high-density lipoprotein cholesterol, low-density lipoprotein cholesterol [LDL-C], total cholesterol, triglycerides, fasting blood glucose, and C-reactive protein) were categorized into tertiles using the empirical distribution of each outcome-specific analytic sample, with the cutoffs reported in the corresponding stratified-analysis table (Tables 3–5).

Table 3.

Stratified associations between childhood loneliness and new-onset hypertension.

Subgroups	n	New-onset hypertension, hazard ratio (95% CI) P-value			P for interaction
		No childhood loneliness	Possible childhood loneliness	Childhood loneliness
Age (yr)					.802
45–55	991	Reference	0.99 (0.77–1.26) .920	1.42 (0.76–2.64) .273
56–62	1028	Reference	1.03 (0.82–1.28) .810	1.54 (1.04–2.30) .032
>63	1045	Reference	0.87 (0.71–1.06) .170	1.04 (0.67–1.60) .866
Marital status					.236
No spouse	537	Reference	0.75 (0.56–1.01) .062	1.28 (0.71–2.31) .420
Married with spouse	2527	Reference	1.01 (0.87–1.16) .940	1.32 (0.98–1.77) .064
Area of residence					.782
Urban	797	Reference	1.03 (0.80–1.32) .823	1.26 (0.67–2.36) .474
Rural	2267	Reference	0.93 (0.80–1.08) .320	1.34 (1.00–1.78) .049
Diabetes					.159
No	2954	Reference	0.96 (0.84–1.09) .519	1.29 (0.99–1.68) .062
Yes	110	Reference	1.26 (0.59–2.68) .544	3.73 (0.39–35.26) .251
Heart disease
No	2814	Reference	0.95 (0.83–1.08) .425	1.33 (1.02–1.74) .038	.991
Yes	250	Reference	1.06 (0.69–1.64) .783	1.08 (0.37–3.17) .894
Arthritis					.907
No	1878	Reference	0.93 (0.79–1.09) .367	1.21 (0.84–1.72) .303
Yes	1180	Reference	1.04 (0.85–1.27) .730	1.38 (0.94–2.03) .105
Dyslipidemia					.352
No	2852	Reference	0.98 (0.86–1.12) .814	1.37 (1.04–1.80) .025
Yes	153	Reference	1.10 (0.62–1.95) .734	0.91 (0.29–2.82) .867
Missing	59	Reference	0.35 (0.09–1.31) .119	0.18 (0.02–1.48) .110
Drinking					.307
Never	2509	Reference	0.98 (0.85–1.13) .811	1.47 (1.12–1.94) .006
Current	424	Reference	0.85 (0.60–1.19) .348	0.67 (0.24–1.86) .439
Ever	131	Reference	0.92 (0.50–1.68) .777	0.36 (0.04–2.99) .345
Smoking					.540
Never	2772	Reference	0.94 (0.83–1.08) .400	1.35 (1.02–1.78) .034
Current	214	Reference	0.93 (0.56–1.56) .792	0.88 (0.32–2.40) .799
Ever	78	Reference	1.75 (0.77–4.02) .184	1.98 (0.32–12.27) .464
Annual income					.069
<15,000	1236	Reference	0.99 (0.81–1.21) .915	1.79 (1.25–2.54) .001
15,000–25,000	242	Reference	1.68 (1.04–2.71) .035	1.00 (0.23–4.47) .995
≥25,000	558	Reference	0.92 (0.68–1.24) .577	1.12 (0.56–2.25) .753
Missing	1028	Reference	0.86 (0.69–1.08) .192	0.90 (0.53–1.52) .689
Depressive symptoms					.959
No	1507	Reference	0.95 (0.79–1.15) .615	1.27 (0.82–1.96) .282
Yes	1400	Reference	0.98 (0.81–1.18) .819	1.25 (0.88–1.77) .220
BMI					.044
Underweight < 18.5	267	Reference	1.05 (0.61–1.79) .870	0.67 (0.19–2.36) .531
Normal weight 18.5–24	1413	Reference	1.13 (0.93–1.38) .224	1.10 (0.69–1.74) .686
Overweight or obesity > 24	942	Reference	0.87 (0.71–1.07) .181	1.50 (1.01–2.25) .047
Missing	442	Reference	0.75 (0.52–1.08) .121	1.88 (1.01–3.50) .048
Fasting blood glucose (mg/dL)					.460
<96.13	742	Reference	1.14 (0.85–1.52) .396	1.14 (0.61–2.12) .683
96.13–107.10	754	Reference	0.82 (0.63–1.06) .126	1.48 (0.94–2.32) .089
>107.10	753	Reference	1.05 (0.83–1.32) .673	1.55 (0.89–2.71) .123
Total cholesterol (mg/dL)					.234
<178.61	741	Reference	1.26 (0.96–1.65) .101	1.14 (0.59–2.22) .696
178.61–210.70	756	Reference	0.85 (0.65–1.10) .205	1.54 (0.95–2.51) .083
>210.70	756	Reference	0.93 (0.73–1.20) .591	1.36 (0.84–2.21) .217
Triglyceride (mg/dL)					.269
<85.8	741	Reference	1.09 (0.81–1.47) .577	1.47 (0.79–2.75) .221
85.8–132.8	759	Reference	0.92 (0.72–1.18) .501	0.90 (0.53–1.53) .694
>132.8	753	Reference	0.93 (0.74–1.18) .549	1.89 (1.19–3.00) .007
HDL-C (mg/dL)					.495
<45.24	737	Reference	0.84 (0.66–1.06) .147	1.14 (0.67–1.93) .627
45.24–57.61	763	Reference	1.00 (0.78–1.28) .991	1.41 (0.86–2.32) .174
>57.61	754	Reference	1.21 (0.91–1.61) .185	1.65 (0.93–2.91) .084
LDL-C (mg/dL)					.554
<103.61	743	Reference	1.17 (0.89–1.52) .259	1.32 (0.73–2.38) .363
103.61–131.06	752	Reference	0.91 (0.70–1.17) .448	1.45 (0.85–2.48) .175
>131.06	756	Reference	0.88 (0.69–1.13) .322	1.40 (0.87–2.25) .161
C-reactive protein (mg/dL)					.073
<0.59	733	Reference	1.29 (0.97–1.70) .079	1.21 (0.68–2.16) .525
0.59–1.41	769	Reference	0.79 (0.61–1.01) .064	1.22 (0.68–2.18) .509
>1.41	751	Reference	0.97 (0.76–1.23) .790	1.81 (1.15–2.87) .011
Disability					.607
No	2457	Reference	0.98 (0.85–1.13) .817	1.32 (0.97–1.80) .075
Yes	605	Reference	0.83 (0.62–1.11) .213	1.30 (0.79–2.16) .304

Above model adjusted for age, marital status, area of residence, drinking, smoking, annual income, BMI, disability. In each case, the model is not adjusted for the stratification variable.

BMI = body mass index, CI = confidence interval, HDL-C = high-density lipoprotein cholesterol, HR = hazard ratios, LDL-C = low-density lipoprotein cholesterol.

2.5. Missing data handling

A complete-case analysis was employed as the primary analytical approach. Childhood loneliness data were ascertained during cohort selection; participants with missing data (n = 747) were excluded before the final analytic sample of 4341 was determined. For participants with missing data on BMI and annual household income, missing-indicator categories were retained in the Cox models to preserve sample size. For other covariates with missingness below 5%, list-wise exclusion was applied, resulting in slightly different sample sizes across models. We elected to use complete-case analysis rather than multiple imputation for several reasons: the proportion of missing covariate data was relatively low for most variables (Table 1): missingness was <5% for marital status, smoking, drinking, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, total cholesterol, triglycerides, fasting blood glucose, and disability; the proportion of missing annual household income was higher (32.1%) and that of body mass index was 15.4%, but these missing-data categories were retained as separate indicator variables in all Cox models (Table 1, footnote); the primary exposure (childhood loneliness) and outcomes (hypertension, heart disease, stroke) were fully ascertained for all included participants; and complete-case analysis can yield unbiased estimates under Missing Completely at Random (MCAR), or under Missing at Random (MAR) conditional on the observed covariates included in our Cox models, which is a reasonable assumption for this cohort. As a sensitivity analysis, we excluded participants with <2 years of follow-up to address potential reverse causation, and the results remained consistent (Table S1, Supplemental Digital Content 2).

Table 1.

Baseline characteristics and new-onset heart disease, hypertension, and stroke by childhood loneliness (China Health and Retirement Longitudinal Study 2011–2020).

Childhood loneliness		All participants		No childhood loneliness		Possible childhood loneliness		Childhood loneliness	P-value
N		4341		1793		2320		228
Age (yr)		61.21 ± 9.63		60.83 ± 10.30		61.39 ± 9.11		62.27 ± 9.19	.041
Fasting blood glucose (mg/dL)		110.22 ± 37.10		110.28 ± 38.41		110.21 ± 36.04		109.76 ± 38.17	.985
Total cholesterol (mg/dL)		199.46 ± 39.21		199.38 ± 38.48		199.15 ± 39.81		203.34 ± 38.32	.418
HDL-C (mg/dL)		51.76 ± 14.65		51.59 ± 14.73		51.79 ± 14.45		52.77 ± 16.05	.615
LDL-C (mg/dL)		120.70 ± 35.49		121.11 ± 35.32		120.09 ± 35.75		124.03 ± 33.86	.340
Triglyceride (mg/dL)		113.28 (80.54–161.96)		110.62 (78.76–161.07)		114.17 (81.42–162.84)		114.17 (83.19–173.68)	.155
C-reactive protein (mg/dL)		1.01 (0.53–2.18)		1.00 (0.54–2.23)		1.03 (0.54–2.18)		0.91 (0.47–1.89)	.389
Marital status									.531
No spouse		828 (19.07%)		355 (19.80%)		428 (18.45%)		45 (19.74%)
Married with spouse		3513 (80.93%)		1438 (80.20%)		1892 (81.55%)		183 (80.26%)
Area of residence									<.001
Urban		1175 (27.07%)		532 (29.67%)		599 (25.82%)		44 (19.30%)
Rural		3166 (72.93%)		1261 (70.33%)		1721 (74.18%)		184 (80.70%)
Dyslipidemia									.954
No		3872 (89.20%)		1592 (88.79%)		2077 (89.53%)		203 (89.04%)
Yes		360 (8.29%)		153 (8.53%)		188 (8.10%)		19 (8.33%)
Missing		109 (2.51%)		48 (2.68%)		55 (2.37%)		6 (2.63%)
Lung disease									.057
No		3954 (91.09%)		1653 (92.19%)		2101 (90.56%)		200 (87.72%)
Yes		375 (8.64%)		137 (7.64%)		212 (9.14%)		26 (11.40%)
Missing		12 (0.28%)		3 (0.17%)		7 (0.30%)		2 (0.88%)
Heart disease									.735
No		3791 (87.33%)		1555 (86.73%)		2037 (87.80%)		199 (87.28%)
Yes		530 (12.21%)		230 (12.83%)		273 (11.77%)		27 (11.84%)
Missing		20 (0.46%)		8 (0.45%)		10 (0.43%)		2 (0.88%)
Drinking									.318
Never		3582 (82.52%)		1491 (83.16%)		1892 (81.55%)		199 (87.28%)
Current		535 (12.32%)		214 (11.94%)		299 (12.89%)		22 (9.65%)
Ever		219 (5.04%)		87 (4.85%)		125 (5.39%)		7 (3.07%)
Missing		5 (0.12%)		1 (0.06%)		4 (0.17%)		0 (0.00%)
Smoking									.550
Never		3923 (90.37%)		1613 (89.96%)		2106 (90.78%)		204 (89.47%)
Current		301 (6.93%)		128 (7.14%)		158 (6.81%)		15 (6.58%)
Ever		100 (2.30%)		44 (2.45%)		47 (2.03%)		9 (3.95%)
Missing		17 (0.39%)		8 (0.45%)		9 (0.39%)		0 (0.00%)
Annual income groups									<.001
<15,000		1850 (42.62%)		699 (38.98%)		1042 (44.91%)		109 (47.81%)
15,000–25,000		345 (7.95%)		144 (8.03%)		190 (8.19%)		11 (4.82%)
≥25,000		753 (17.35%)		346 (19.30%)		364 (15.69%)		43 (18.86%)
Missing		1393 (32.09%)		604 (33.69%)		724 (31.21%)		65 (28.51%)
Depressive symptoms									<.001
No		2070 (50.25%)		873 (51.96%)		1113 (50.18%)		84 (38.01%)
Yes		2049 (49.75%)		807 (48.04%)		1105 (49.82%)		137 (61.99%)
BMI groups									.001
Underweight < 18.5		321 (7.39%)		142 (7.92%)		152 (6.55%)		27 (11.84%)
Normal weight 18.5–24		1818 (41.88%)		704 (39.26%)		1019 (43.92%)		95 (41.67%)
Overweight or obesity ≥24		1533 (35.31%)		636 (35.47%)		822 (35.43%)		75 (32.89%)
Missing	669 (15.41%)		311 (17.35%)		327 (14.09%)		31 (13.60%)
Disability									.127
No	3431 (79.16%)		1442 (80.60%)		1814 (78.29%)		175 (76.75%)
Yes	903 (20.84%)		347 (19.40%)		503 (21.71%)		53 (23.25%)
Childhood loneliness									<.001
No childhood loneliness	1793 (41.30%)		1793 (100.00%)		0 (0.00%)		0 (0.00%)
Possible childhood loneliness	2320 (53.44%)		0 (0.00%)		2320 (100.00%)		0 (0.00%)
Childhood loneliness	228 (5.25%)		0 (0.00%)		0 (0.00%)		228 (100.00%)
New-onset heart disease									.012
No	2938 (68.63%)		1249 (70.97%)		1545 (67.32%)		144 (63.72%)
Yes	1343 (31.37%)		511 (29.03%)		750 (32.68%)		82 (36.28%)
New-onset hypertension									.050
No	2036 (47.56%)		859 (48.81%)		1086 (47.32%)		91 (40.27%)
Yes	2245 (52.44%)		901 (51.19%)		1209 (52.68%)		135 (59.73%)
New-onset stroke									.034
No	3867 (89.08%)		1622 (90.46%)		2048 (88.28%)		197 (86.40%)
Yes	474 (10.92%)		171 (9.54%)		272 (11.72%)		31 (13.60%)

Continuous variables presented as mean ± SD or Median (interquartile range); categorical variables as percentage (%); P-values from t test, Mann–Whitney U test, or chi-square test as appropriate.

BMI = body mass index, CI = confidence interval, HDL-C = high-density lipoprotein cholesterol, HR = hazard ratios, LDL-C = low-density lipoprotein cholesterol, SD = standard deviation.

2.6. Statistical analysis

We first described baseline characteristics of the study sample across the 3 mutually exclusive childhood-loneliness groups (no, possible, and definite childhood loneliness). Continuous variables were presented as mean ± standard deviation (SD) when approximately normally distributed and as median (interquartile range) otherwise; categorical variables were presented as counts and percentages. Normality was assessed by the Shapiro–Wilk test and inspection of quantile–quantile plots. Group differences were evaluated with 1-way analysis of variance or the Kruskal–Wallis test for continuous variables, and Pearson chi-square test or Fisher exact test for categorical variables, as appropriate. For each outcome (hypertension, heart disease, stroke), we constructed Kaplan–Meier curves and used log-rank tests to compare event-free survival across the 3 childhood-loneliness groups. We then fitted cause-specific Cox proportional hazards regression models to estimate the association between childhood loneliness and incident outcomes. To balance confounding control against overadjustment for mediators, we adopted a 3-model approach. Model 1 was unadjusted. Model 2 (the primary confounder model) adjusted for baseline demographic and socioeconomic factors: age (continuous, years), marital status (married or cohabiting vs not), area of residence (rural vs urban), current smoking (yes vs no), current alcohol drinking (yes vs no), and annual household income (quartiles). Model 3 additionally adjusted for BMI groups (underweight < 18.5, normal weight 18.5–<24.0, overweight/obese ≥ 24.0 kg/m²; missing-category indicator retained) and disability status, both of which may lie on the causal pathway between childhood loneliness and adult CVD. Time-on-study, defined as the elapsed time from each participant’s 2011 baseline interview to the first of: incident cardiovascular event, death from any cause, loss to follow-up, or the end of the 2020 wave, was used as the primary time scale in all 3 outcome models (i.e., a uniform approach across hypertension, heart disease, and stroke). This choice yields effect estimates that are robust to calendar-period effects and is consistent with the standard CHARLS analytic approach. Participants contributed person-time from baseline until the first of: the incident event of interest, death from any cause, loss to follow-up, or the administrative end of follow-up at the 2020 CHARLS wave. Age at study entry was included as a continuous covariate in all Cox models rather than being used to define a separate time scale, in order to provide a uniform approach across the 3 outcomes. The proportional hazards assumption was checked using scaled Schoenfeld residuals (Grambsch–Therneau global test and covariate-specific tests), supplemented by visual inspection of log(–log(survival)) plots. No significant violations were detected for the exposure variable in any model (all global-test P > .05). Hazard ratios (HRs) with 95% confidence intervals (CIs) were calculated. In the main results, we report multivariable-adjusted HRs comparing the possible-loneliness and childhood loneliness groups to the no-loneliness reference group. All tests were 2-sided, with the significance threshold set at α = 0.05. Because hypertension, heart disease, and stroke are biologically distinct co-primary endpoints, no correction for multiple comparisons was applied to the 3 primary effect estimates.

The primary analysis used a complete-case approach under the assumption that covariate data are MAR conditional on the observed variables in the Cox models; covariate missingness was below 5% for the majority of variables (see Table 1, footnote, for a detailed breakdown). No multiple imputation was performed as the primary analysis; however, to assess robustness to the missing-data assumption, we conducted a multiple-imputation sensitivity analysis. Missing values in BMI, annual household income, smoking, drinking, and disability (the covariates with any missingness) were imputed by fully conditional specification (chained equations) across M = 20 datasets, with the exposure, outcomes, and all covariates included in the imputation model. Cox models were refitted in each imputed dataset, and estimates were pooled by Rubin rules. The primary exposure and outcomes were fully ascertained and therefore required no imputation. We conducted stratified analyses for each outcome across the following prespecified baseline variables: age tertiles (for hypertension: 45–55, 56–62, >63 years; for heart disease and stroke: 45-55, 56–63, >63 years, based on the tertile cutoffs of the respective analytic sample distribution), area of residence (rural vs urban), marital status, current smoking, current alcohol drinking, annual household income quartile (categorical: <15,000; 15,000–24,999; ≥25,000 CNY), BMI category, depressive symptoms, dyslipidaemia, diabetes mellitus, arthritis, high-density lipoprotein cholesterol (HDL-C) stratum, and C-reactive protein (CRP) stratum. Each stratum-specific model excluded the stratification variable from the covariate set. Effect modification was formally tested by multiplicative interaction terms (Wald test, α = 0.05). Given the exploratory nature of these subgroup analyses, findings were interpreted with caution and without correction for multiple comparisons. We performed several sensitivity analyses to assess the robustness of our findings. First, we repeated all Cox analyses after excluding participants who had <2 years of follow-up, to address potential reverse causation (Table S1, Supplemental Digital Content 2). Second, we reclassified the exposure into an alternative binary variable in which participants meeting the 2 childhood-loneliness criteria were grouped as exposed versus those meeting neither or participants meeting at least 1 of the 2 childhood-loneliness criteria to evaluate the impact of the strict exposure definition (Table S2, Supplemental Digital Content 1). Third, we refitted model 3 after additionally adjusting for baseline depressive symptoms to assess whether adult mental health attenuated the observed associations (Table S3, Supplemental Digital Content 3). All statistical analyses were performed using EmpowerStats version 4.1 (http://www.empowerstats.com; X&Y Solutions, Inc, Boston), a statistical platform built on R software version 4.2.0. XC is employed by X&Y Solutions, Inc; this affiliation is disclosed in the author contributions and competing interests sections (http://www.r-project.org; R Foundation for Statistical Computing, Vienna, Austria), with the survival and ggpubr packages for Cox modeling and visualization.

3. Results

3.1. Baseline characteristics

In Table 1, a total of 4341 women with low educational attainment (mean age 61.21 years) were included. The majority were rural residents (72.93%) and married (80.93%). The prevalence of current smoking (6.93%) and drinking (12.32%) was low. Regarding childhood loneliness, 228 (5.25%) reported often feeling lonely and having no close friend, 2320 (53.44%) met 1 criterion (possible loneliness), and 1793 (41.30%) reported no loneliness. Baseline characteristics including age, residence, income, depressive symptoms, and BMI differed according to loneliness status (all P < .05). The incidence of new-onset heart disease, hypertension, and stroke was highest among those with childhood loneliness (all P ≤ .05).

3.2. Incidence of outcomes

These differences reached nominal statistical significance in chi-square tests of cumulative incidence across the 3 exposure categories (Table 1): P = .050 for hypertension, P = .012 for heart disease, and P = .034 for stroke. The corresponding multivariable-adjusted cumulative incidence curves (Fig. 2) yielded Cox proportional hazards model-derived P-values of P = .042 for hypertension, P = .020 for heart disease, and P = .067 for stroke, all consistent with the Wald-test P-values reported in Table 2, preserving the qualitative pattern in which the childhood loneliness group exhibited the steepest cumulative event trajectory across all 3 outcomes.

3.3. Association of childhood loneliness with outcomes

The associations between childhood loneliness and the incidence of heart disease, hypertension, and stroke were examined using Cox proportional hazards models, with results presented in Table 2. In the fully adjusted models (model 3), which controlled for a comprehensive set of demographic, socioeconomic, and health-related covariates, childhood loneliness was independently associated with both new-onset heart disease and hypertension. Specifically, individuals reporting childhood loneliness exhibited a 41% higher hazard of developing heart disease compared to their counterparts who did not experience loneliness (HR = 1.41; 95% CI: 1.06–1.89; P = .020), and a statistically significant association was also observed for hypertension, with childhood loneliness associated with a 31% increased risk (HR = 1.31; 95% CI: 1.01–1.70; P = .042). In contrast, no statistically significant association was found between childhood loneliness and the incidence of stroke, as the point estimate suggested an elevated risk but did not reach statistical significance in the fully adjusted model (HR = 1.47, 95% CI: 0.97–2.21, P = .067).

3.4. Subgroup analyses

We conducted stratified analyses and formal interaction tests to evaluate whether the association between childhood loneliness and incident cardiovascular outcomes varied by baseline characteristics (Tables 3–5). For new-onset hypertension (Table 3), the positive association between childhood loneliness and increased risk appeared more pronounced in certain subgroups. Specifically, elevated hazard ratios were observed among participants aged 56 to 62 years (HR = 1.54; 95% CI: 1.04–2.30), those living in rural areas (HR = 1.34; 95% CI: 1.00–1.78), individuals without dyslipidemia (HR = 1.37; 95% CI: 1.04–1.80), and never-drinkers (HR = 1.47; 95% CI: 1.12–1.94). The association was also stronger among participants with an annual income below 15,000 (HR = 1.79; 95% CI: 1.25–2.54). However, despite variation in point estimates across strata, tests for interaction were not statistically significant for most variables (all P for interaction > 0.05), with the exception of BMI, which suggested a potential interaction (P for interaction = 0.044).

For new-onset heart disease (Table 4), the association between childhood loneliness and new-onset heart disease was significantly modified by baseline depressive symptoms (P for interaction = 0.007). A significant positive association was observed only among participants with depressive symptoms (HR = 1.67; 95% CI: 1.16–2.42), whereas no significant association was observed among participants without depressive symptoms (HR = 0.89; 95% CI: 0.51–1.55). The association appeared stronger in several other subgroups, including individuals with normal BMI (HR = 2.05; 95% CI: 1.33–3.16), never-smokers (HR = 1.38; 95% CI: 1.01–1.88), those with higher baseline high-density lipoprotein cholesterol (HDL-C > 56.45 mg/dL; HR = 2.01; 95% CI: 1.21–3.34), and those with lower C-reactive protein (CRP < 0.64 mg/dL; HR = 2.50; 95% CI: 1.49–4.20). However, interaction tests for these variables were not statistically significant.

Table 4.

Stratified associations between childhood loneliness and new-onset heart disease.

Subgroups	n	New-onset heart disease, hazard ratio (95% CI) P-value			P for interaction
		No childhood loneliness	Possible childhood loneliness	Childhood loneliness
Age (yr)					.824
45–55	1119	Reference	1.23 (0.91–1.66) .177	1.03 (0.44–2.37) .951
56–63	1344	Reference	1.11 (0.88–1.40) .394	1.23 (0.78–1.94) .365
>63	1277	Reference	1.17 (0.91–1.50) .227	1.58 (1.02–2.45) .039
Marital status					.282
No spouse	697	Reference	1.49 (1.05–2.13) .027	1.48 (0.72–3.04) .281
Married with spouse	3043	Reference	1.11 (0.95–1.31) .192	1.38 (1.01–1.90) .046
Area of residence					.602
Urban	977	Reference	1.14 (0.86–1.52) .365	1.03 (0.51–2.06) .941
Rural	2763	Reference	1.21 (1.01–1.43) .035	1.51 (1.09–2.08) .012
Diabetes					.438
No	3563	Reference	1.17 (1.01–1.37) .043	1.37 (1.01–1.85) .040
Yes	177	Reference	1.34 (0.73–2.44) .341	3.31 (0.94–11.64) .062
Arthritis					.969
No	2285	Reference	1.17 (0.96–1.42) .130	1.39 (0.93–2.10) .112
Yes	1455	Reference	1.21 (0.96–1.51) .103	1.41 (0.93–2.13) .108
Dyslipidemia					.063
No	3399	Reference	1.26 (1.08–1.48) .004	1.51 (1.11–2.07) .010
Yes	250	Reference	0.68 (0.45–1.05) .082	1.42 (0.59–3.43) .433
Missing	91	Reference	1.22 (0.31–4.83) .775	0.00 (0.00–Inf) .998
Drinking					.629
Never	3055	Reference	1.13 (0.96–1.32) .145	1.40 (1.03–1.91) .032
Current	492	Reference	1.51 (0.91–2.48) .109	1.02 (0.30–3.49) .975
Ever	193	Reference	1.33 (0.68–2.59) .401	2.41 (0.60–9.63) .215
Smoking					.155
Never	3411	Reference	1.21 (1.03–1.41) .018	1.38 (1.01–1.88) .045
Current	247	Reference	0.74 (0.41–1.34) .323	1.42 (0.46–4.43) .541
Ever	82	Reference	2.75 (1.06–7.11) .037	4.82 (1.25–18.55) .022
Annual income					.274
<15,000	1576	Reference	1.42 (1.13–1.80) .003	1.82 (1.23–2.71) .003
15,000–25,000	285	Reference	1.47 (0.87–2.47) .150	1.27 (0.29–5.56) .754
≥25,000	651	Reference	0.87 (0.62–1.23) .422	0.94 (0.43–2.07) .879
Missing	1228	Reference	1.08 (0.82–1.42) .582	1.22 (0.68–2.21) .502
Depressive symptoms					.007
No	1858	Reference	0.91 (0.73–1.14) .413	0.89 (0.51–1.55) .689
Yes	1693	Reference	1.45 (1.17–1.80) .001	1.67 (1.16–2.42) .006
BMI					.244
Underweight < 18.5	293	Reference	1.01 (0.57–1.78) .971	0.99 (0.34–2.94) .990
Normal weight 18.5–24	1632	Reference	1.36 (1.06–1.73) .015	2.05 (1.33–3.16) .001
Overweight or obesity ≥24	1284	Reference	1.00 (0.79–1.25) .979	0.96 (0.57–1.62) .866
Missing	531	Reference	1.45 (0.97–2.18) .072	1.88 (0.85–4.12) .118
Fasting blood glucose (mg/dL)					.758
<96.84	922	Reference	1.45 (1.06–1.98) .021	1.65 (0.93–2.93) .088
96.84–108.19	924	Reference	1.14 (0.84–1.54) .392	1.64 (0.94–2.83) .079
>108.19	920	Reference	1.29 (0.96–1.72) .088	1.09 (0.56–2.12) .809
Total cholesterol (mg/dL)					.946
<179.77	909	Reference	1.44 (1.04–1.99) .029	1.96 (1.04–3.68) .037
179.77–212.25	933	Reference	1.19 (0.90–1.58) .227	1.48 (0.85–2.60) .169
>212.25	929	Reference	1.28 (0.94–1.73) .116	1.35 (0.75–2.42) .319
Triglyceride (mg/dL)					0.725
<88.5	914	Reference	1.29 (0.94–1.77) .120	2.09 (1.17–3.71) .012
88.5–137.18	928	Reference	1.36 (1.00–1.84) .051	1.29 (0.69–2.40) .429
>137.18	930	Reference	1.11 (0.84–1.47) .469	1.17 (0.65–2.08) .602
HDL-C (mg/dL)					.307
<44.46	905	Reference	0.97 (0.72–1.31) .843	0.95 (0.47–1.91) .890
44.46–56.45	924	Reference	1.39 (1.03–1.86) .031	1.41 (0.75–2.62) .284
>56.45	946	Reference	1.36 (1.00–1.84) .047	2.01 (1.21–3.34) .007
LDL-C (mg/dL)					.922
<103.61	903	Reference	1.25 (0.92–1.71) .154	1.57 (0.84–2.95) .161
103.61–132.70	935	Reference	1.27 (0.94–1.71) .117	1.17 (0.64–2.13) .612
>132.70	931	Reference	1.24 (0.92–1.67) .162	1.62 (0.93–2.81) .086
C-reactive protein (mg/dL)					.174
<0.64	923	Reference	1.48 (1.06–2.07) .021	2.50 (1.49–4.20) .001
0.64–1.54	927	Reference	1.16 (0.87–1.54) .312	0.91 (0.47–1.77) .781
>1.54	924	Reference	1.12 (0.84–1.50) .441	1.13 (0.58–2.21) .715
Disability					.553
No	2977	Reference	1.15 (0.98–1.36) .094	1.49 (1.08–2.08) .016
Yes	758	Reference	1.32 (0.94–1.85) .111	1.23 (0.65–2.33) .515

Above model adjusted for age, marital status, area of residence, drinking, smoking, annual income, BMI, disability. In each case, the model is not adjusted for the stratification variable.

BMI = body mass index, CI = confidence, HDL-C = high-density lipoprotein cholesterol, HR = hazard ratios, LDL-C = low-density lipoprotein cholesterol.

For stroke (Table 5), the strongest association with childhood loneliness was observed among participants with arthritis (HR = 1.85; 95% CI: 1.06–3.25), current smokers (HR = 3.44; 95% CI: 1.03–11.46), and those with higher CRP levels above 1.59 mg/dL, with a hazard ratio of 2.11 (95% CI: 1.07–4.16). However, none of the interaction tests reached statistical significance (all P for interaction > .05), indicating that the relationship between childhood loneliness and stroke risk was not significantly different across the examined subgroups.

3.5. Sensitivity analyses

To assess the robustness of our primary findings and address potential concerns about reverse causality, we conducted a sensitivity analysis excluding participants with a follow-up duration of less than 2 years. As detailed in Table S1, Supplemental Digital Content 2, the results remained generally consistent with the main analysis across all outcomes. Specifically, in the fully adjusted model (Model 3), the association between childhood loneliness and new-onset hypertension (HR = 1.32, 95% CI: 1.02–1.72) and heart disease (HR = 1.40, 95% CI: 1.04–1.88) persisted, remaining statistically significant. For new-onset stroke, the point estimate (HR = 1.47, 95% CI: 0.97–2.21) was nearly identical to that in the main analysis. These findings suggest that the inclusion of participants with very short follow-up did not introduce significant bias, thereby reinforcing the robustness of our primary conclusions. In the alternative binary classification (Table S2, Supplemental Digital Content 1), childhood loneliness (meeting both criteria vs meeting neither or one) remained associated with new-onset hypertension (HR = 1.34, 95% CI: 1.04–1.72, P = .021); the associations with heart disease (HR = 1.27, 95% CI: 0.97–1.67, P = .087) and stroke (HR = 1.34, 95% CI: 0.91–1.96, P = .142) were attenuated and did not reach statistical significance. When Model 3 was additionally adjusted for baseline depressive symptoms (Table S3, Supplemental Digital Content 3), the associations were further attenuated: heart disease (HR = 1.32, 95% CI: 0.98–1.79, P = .067), hypertension (HR = 1.25, 95% CI: 0.95–1.64, P = .111), and stroke (HR = 1.41, 95% CI: 0.93–2.15, P = .108), suggesting that depressive symptoms may partially mediate the observed associations. In a multiple-imputation sensitivity analysis (M = 20, fully conditional specification), the pooled associations for definite childhood loneliness were essentially unchanged and remained directionally consistent (heart disease: HR = 1.40, 95% CI: 0.93–2.10; hypertension: HR = 1.31, 95% CI: 0.93–1.84; stroke: HR = 1.46, 95% CI: 0.80–2.66), confirming that the complete-case estimates are not materially biased by missing covariate data (Table S4, Supplemental Digital Content 4).

4. Discussion

In this 9-year prospective study of middle-aged and older Chinese women with low educational attainment, we found that those who reported experiencing loneliness in childhood had a significantly higher risk of developing hypertension and heart disease in later life. To our knowledge, this is the first study to demonstrate a prospective association between childhood loneliness and incident major cardiovascular outcomes. Women who experienced childhood loneliness had a 31% higher hazard of hypertension and a 41% higher hazard of heart disease after adjusting for a comprehensive set of adult risk factors. Although the association with stroke was elevated by approximately 47%, it did not reach statistical significance, possibly due to the smaller number of stroke events and consequent limited statistical power.

It is noteworthy that the present analysis focused specifically on low-educated women, a vulnerable subpopulation not separately analyzed in the recent nationwide CHARLS-based study by Long and Li,^[10] which examined the full CHARLS cohort of 12,431 middle-aged and older adults. Our findings, therefore, complement rather than replicate the prior report by examining effect modification within a high-risk subgroup for which life-course cardiovascular risk factors may be particularly salient. Our findings align with and extend the existing literature on early-life psychosocial adversity and cardiovascular health. The magnitude of the associations we observed (HRs ~1.3–1.5) is comparable to those reported for more conventional measures of childhood adversity. A scientific statement from the American Heart Association concluded that childhood adversity is associated with a 30% to 60% higher risk of cardiometabolic diseases, including heart disease and stroke.^[6] Meta-analyses have similarly shown that individuals exposed to adverse childhood experiences (ACEs) have significantly higher rates of ischemic heart disease, with even a single adversity conferring a 30% or greater increase in risk^[7,21] Our results specifically highlight loneliness, a subjective emotional experience, as a distinct form of adversity that carries similar long-term cardiovascular consequences. Our observed effect sizes are consistent with previous research on adult social isolation and loneliness. In a meta-analysis by Valtorta et al., loneliness and social isolation in adulthood were associated with a 29% increased risk of coronary heart disease and a 32% increased risk of stroke.^[3] Our finding of a 41% increased risk for heart disease and a nonsignificant 47% increased risk for stroke associated with childhood loneliness closely parallels these estimates, suggesting that the cardiovascular risk associated with loneliness may persist across the life course, independent of adulthood social isolation. For hypertension, our results echo findings from the Nurses’ Health Study II, where severe childhood abuse was associated with approximately 20% higher hypertension incidence after adjusting for potential mediators,^[8] and from a recent meta-analysis reporting that various ACEs collectively contribute to higher odds of hypertension in adulthood.^[22] Thus, the burden of evidence suggests that early psychosocial stressors – whether abuse, neglect, or as our study highlights, loneliness – have measurable associations with cardiovascular risk.

Our focus on women with low educational attainment is supported by prior evidence suggesting sex-specific and socioeconomic differences in the health outcomes associated with childhood adversity. Women are more likely to report certain interpersonal adversities and may exhibit greater physiological responsiveness to social stress in terms of CVD risk.^[16,23] Previous research found that childhood adversity was associated with CVD in women but not in men, with women experiencing high adversity and additional adult stressors showing particularly elevated risk.^[10] In our all-female sample, childhood loneliness demonstrated significant cardiovascular associations. Sex-specific analyses (e.g., parallel analyses in CHARLS-based male samples) were beyond the scope of the present study and are warranted to formally test whether the present associations differ by sex. Low educational attainment – and correlated socioeconomic disadvantage – may amplify the health outcomes associated with childhood adversity by introducing additional stressors and limiting access to resources that could buffer early-life insults.^[24] In our cohort, women with childhood loneliness predominantly grew up in rural, economically modest conditions and remained in low socioeconomic strata as adults, potentially compounding the physiological imprint of early adversity.

The significant association between childhood loneliness and incident hypertension merits particular attention. Hypertension is a key mediator in the pathway to heart disease and stroke, with well-established genetic and lifestyle determinants. Our data suggest that beyond traditional risk factors such as obesity, smoking, and diet, the psychosocial environment in childhood is associated with the development of elevated blood pressure decades later. This aligns with life-course studies demonstrating that early-life psychosocial stress is associated with adverse blood pressure trajectories.^[25] A longitudinal analysis found that individuals with multiple adverse childhood experiences exhibited greater increases in blood pressure from childhood into young adulthood compared to those without such exposures.^[25] Research on loneliness in older adults has similarly shown that loneliness is associated with increases in systolic blood pressure over time, with effects accumulating year by year and remaining significant after adjusting for baseline blood pressure and conventional risk factors.^[26] Our study extends this concept backward in time, suggesting that loneliness experienced during childhood may be associated with pathophysiological processes that manifest as hypertension much later in life.

Several plausible biological and behavioral mechanisms may explain the observed associations. Chronic stress and hypothalamic-pituitary-adrenal axis dysregulation represent a key pathway. Childhood loneliness likely constitutes a form of chronic interpersonal stress – a child who frequently feels alone or unsupported may experience ongoing anxiety, sadness, and lack of emotional security. Such chronic stress can become biologically embedded in regulatory systems. Research on early-life adversity indicates that these exposures are associated with persistent hyperactivation of the hypothalamic-pituitary-adrenal axis and sympathetic nervous system.^[27,28] Children exposed to emotional adversity have been found to exhibit elevated cortisol levels and altered diurnal cortisol patterns years after the exposure.^[29,30] Loneliness is accompanied by feelings of threat and hypervigilance, which can acutely increase vascular resistance and blood pressure.^[31] Over the long term, repeated or prolonged activation of stress pathways – including elevated cortisol and catecholamines – may be associated with hypertension through arterial stiffness, renal sodium retention, and metabolic dysregulation.

Behavioral pathways likely play a mediating role. Individuals who experience early-life adversity may develop maladaptive health behaviors over time. Lonely children or adolescents may be more prone to smoking, excessive alcohol use, poor dietary habits, or physical inactivity as they age, potentially as coping mechanisms or due to social skill deficits.^[32,33] In our analysis, adjusting for baseline smoking and BMI modestly attenuated the associations, suggesting partial mediation through these behavioral factors. Prior research has consistently linked adverse childhood experiences with unhealthy behaviors, including smoking initiation in adolescence, heavy drinking, obesity, and poor sleep quality.^[34,35] Loneliness specifically has been associated with lower physical activity levels and poorer adherence to medical advice in adulthood, which may further worsen cardiovascular risk profiles.

Mental health represents another important pathway. Chronic loneliness can precipitate depression and anxiety, which are themselves established risk factors for coronary heart disease and stroke through both behavioral mechanisms (e.g., physical inactivity, poor medication adherence) and direct physiological effects (inflammation, autonomic dysfunction).^[36] In our data, we observed that women with childhood loneliness had higher rates of depressive symptoms in adulthood (data not shown), consistent with previous CHARLS analyses.^[14] While we did not adjust for depression in primary models due to concerns about overadjustment and potential reverse causation, prior studies have found that depression partially mediates the relationship between childhood adversities and CVD.^[16]

Inflammation and immune programming may also contribute. Research has demonstrated that childhood trauma or chronic stress is associated with a pro-inflammatory phenotype in later life.^[37] Elevated levels of inflammatory biomarkers, including C-reactive protein and interleukin-6, have been observed in adults with adverse childhood environments, even decades after exposure.^[38] Chronic inflammation promotes atherosclerosis and hypertension, providing a plausible biological link between early loneliness and later CVD. Emerging evidence further implicates endothelial dysfunction. Su and colleagues found that young adults with ACE exposure had significantly higher circulating endothelin-1, a potent vasoconstrictor and marker of endothelial activation.^[25] Endothelin-1 is implicated in hypertension and atherosclerosis development,^[39,40] suggesting that childhood loneliness may upregulate stress-responsive vascular mechanisms that increase cardiovascular susceptibility.

The lack of a statistically significant association with stroke in our study warrants consideration. Despite a point estimate suggesting a 47% increased hazard, the confidence interval was wide and included the null. This may reflect the smaller number of stroke events (n = 474) relative to hypertension and heart disease, limiting statistical power to detect modest associations. Alternatively, stroke in older adults has diverse etiologies and strong competing risk factors. These include hypertension, diabetes, and atrial fibrillation, which may reduce the apparent association with early-life psychosocial factors. Nevertheless, prior studies provide suggestive evidence for this connection. In the meta-analysis by Valtorta et al, social isolation was associated with a 32% higher risk of incident stroke.^[3] The original ACE Study found graded increases in self-reported stroke with higher ACE counts,^[41] and other cohorts have noted that early-life psychosocial stress correlates with greater stroke incidence.^[42] Our findings are directionally consistent with these studies, and we suspect that with greater statistical power, a significant relationship might emerge. The P-value of .067 indicates a trend that warrants confirmation in larger samples or with longer follow-up.

5. Strengths and limitations

Key strengths of this study include its prospective design, the use of a well-characterized national cohort (CHARLS), a relatively long follow-up period (median 9 years), and the ability to control for an extensive set of adult confounders. By focusing on a homogeneous subgroup (low-educated women), we minimized confounding by sex and education. However, several important limitations must be acknowledged. Recall bias is a concern, as our measure of childhood loneliness was retrospective, based on participants’ recollection of feelings from many decades earlier. Individuals with certain adult characteristics, such as current depression or social isolation, may be more prone to recalling or reporting loneliness experienced in childhood, which could potentially inflate associations if those individuals also exhibit higher cardiovascular risk. We attempted to mitigate this by adjusting for baseline factors including marital status (as a proxy for current social support) and by conducting sensitivity analyses excluding participants with less than 2 years of follow-up to reduce bias from incipient illness. Nonetheless, recall bias cannot be entirely eliminated. To further address recall bias, we defined childhood loneliness using a strict criterion requiring both frequent loneliness AND absence of a close friend, which reduces the likelihood of capturing vague recollections. Additionally, we adjusted for baseline depressive symptoms and marital status in sensitivity analyses to partially account for the influence of current psychological state on retrospective recall. Prospective studies measuring loneliness during childhood or adolescence would be valuable to confirm our findings. Residual confounding by unmeasured factors is also possible. Childhood loneliness may indicate other adversities, such as parental neglect, orphanhood, or chronic illness during childhood. These adversities themselves are associated with later CVD. We lacked detailed data on all childhood adversities in the baseline survey. However, our sensitivity analysis in the life history subset suggested that adjusting for childhood economic hardship did not abolish the association. Other unmeasured factors, such as genetic predispositions, childhood nutrition, and community characteristics, were not accounted for. It is important to note that causality cannot be definitively established. For instance, a third factor, such as a difficult temperament or underlying neurodevelopmental condition, may be associated with both loneliness and a heightened risk of CVD. We attempted to address this partially by controlling for disability status and adult health behaviors, but some residual confounding may persist. Furthermore, in the primary analysis we retained a 3-level exposure classification and did NOT combine participants meeting only 1 criterion into the reference group; this avoids the misclassification that would otherwise arise. An alternative binary exposure (meeting both criteria vs meeting neither or 1 criterion) was evaluated as a sensitivity analysis (Table S2, Supplemental Digital Content 1). We elected to use this conservative, high-specificity approach to maximize confidence that the exposed group represents individuals with pronounced childhood loneliness. As a sensitivity analysis, we tested an alternative binary classification with participants meeting both criteria classified as exposed and those meeting neither or 1 criterion classified as the reference group; effect estimates were attenuated relative to the primary 3-level analysis (Table S2, Supplemental Digital Content 1). Outcome ascertainment for heart disease and stroke relied on self-reported physician diagnoses. In resource-limited settings, some events may go undiagnosed or unreported, potentially leading to misclassification. However, such misclassification would likely be nondifferential with respect to childhood loneliness, biasing estimates toward the null and suggesting that our findings may be conservative. Hypertension was measured objectively using standardized clinical assessments, strengthening confidence in the observed associations. Nevertheless, validation studies within CHARLS have demonstrated acceptable reliability of self-reported chronic disease diagnoses.^[19] We did not differentiate subtypes of heart disease (e.g., angina vs myocardial infarction) or stroke (ischemic vs hemorrhagic). A more granular outcome assessment might reveal whether associations are stronger for specific conditions. The focus on low-educated Chinese women as the study population may limit the generalizability of our findings. Cultural and cohort factors could be associated with both experiences of childhood loneliness and cardiovascular risk. These women predominantly grew up in the 1950s-1960s in rural China, an era of significant social upheaval including the Great Chinese Famine and Cultural Revolution. Loneliness in this context may frequently have co-occurred with material hardship. It is possible that in other populations, such as men or individuals in more affluent countries, the magnitude of associations might differ. We encourage caution in extrapolating our effect sizes directly to other groups. Nonetheless, given the fundamental psychosocial and biological processes involved, we expect the qualitative finding that early loneliness can have long-term health associations to be relevant beyond this specific demographic. Future studies should examine whether these associations extend to men, individuals with higher socioeconomic status, and populations in different cultural contexts. Additionally, research employing trajectory analyses of loneliness across multiple life stages would help disentangle the timing of associations and clarify whether critical periods exist during which loneliness exerts particularly strong effects on later cardiovascular health. We lacked data on adult loneliness at baseline, as CHARLS measured current loneliness through a single item within a depression scale. Adult and childhood loneliness are correlated in our sample,^[14] raising the question of whether observed associations are attributable to current rather than childhood loneliness. However, a recent CHARLS analysis found that childhood loneliness was associated with cognitive decline regardless of adult loneliness,^[14] suggesting an independent association of early-life exposure. In our study, we did not adjust for adult loneliness to avoid overadjustment bias. This is because adult loneliness may lie on the causal pathway; childhood loneliness may be associated with adult loneliness, which subsequently affects health outcomes. Our binary measure of childhood loneliness, based on 2 retrospective questions, lacks nuance regarding chronicity, timing, and specific reasons for loneliness. Qualitative research might provide deeper insights into the nature of loneliness these women experienced.

6. Conclusions

In conclusion, this study provides novel evidence that loneliness experienced during childhood is associated with an elevated risk of developing hypertension and heart disease many years later in adulthood. These findings suggest that early-life social and emotional experiences may be associated with CVD risk. These findings highlight the importance of early psychosocial experiences for cardiovascular health and suggest that interventions targeting childhood social well-being may have long-term cardiovascular benefits. Future research is warranted to examine whether these associations extend to men and individuals with higher socioeconomic status.

Acknowledgments

This study used data from the CHARLS. We thank the CHARLS research team for their time and effort on the CHARLS project.

Author contributions

Conceptualization: Yan Wang, Xinglin Chen.

Data curation: Lei Li, Jun Wang.

Formal analysis: Lei Li, Xinglin Chen.

Investigation: Lei Li, Xinglin Chen.

Methodology: Yan Wang, Lei Li, Jun Wang, Xinglin Chen.

Supervision: Yan Wang.

Validation: Yan Wang, Lei Li, Jun Wang, Xinglin Chen.

Writing – original draft: Yan Wang.

Writing – review & editing: Yan Wang, Lei Li.