Associations of deprivation and income-poverty with anxiety and stress among Hong Kong adults: a longitudinal household survey

Abstract

Objectives

To explore the cross-sectional and longitudinal associations of poverty, assessed by deprivation index (DI) and income, with anxiety and stress among adults

Design

Longitudinal cohort observational study

Setting

A cohort of adults established in 2014–2015 (T1) in Hong Kong.

Participants

493 participants (177 men, median age 56 years) were followed up in the third wave of household survey in 2021–2022 (T3).

Main outcome measures

Self-reported equivalised household income and DI, measured as material and social deprivation, were recorded. Anxiety and stress were assessed using the Depression, Anxiety and Stress Scale-21 items (Chinese) version.

Results

Being deprived in T3 is significantly associated with higher anxiety and stress score in T3 as compared with being non-deprived cross-sectionally. Income status did not show significant association with anxiety and stress score cross-sectionally in T3. Increased deprivation status since baseline is significantly associated with higher anxiety and stress score in T3 than persistent non-deprivation. Being able to afford one less basic necessity throughout T1–T3 is significantly associated with 0.31-unit higher anxiety score (95% CI 0.18 to 0.45) and 0.36-unit higher stress score (95% CI 0.18 to 0.55) across time.

Subgroup analyses further showed that income-poor younger adults aged less than 60 years tended to have higher anxiety and stress scores than their non-income-poor counterparts in the same age group in T3 cross-sectionally. Younger adults with occasional non-income-poverty throughout T1–T3 also had higher anxiety and stress scores in T3 than persistently non-income-poor younger adults. This relationship of income-poverty with anxiety and stress was not significant among older adults aged 60 years or above.

Conclusions

Individual deprivation and increased deprivation over time were associated with higher anxiety and stress scores in T3 cross-sectionally and longitudinally respectively. Young adults’ anxiety and stress scores tended to be more affected by income-poverty than their older counterparts. Nevertheless, sample attrition may occur as a proportion of participants dropped out throughout the 7 years of follow-up.

Age-specific poverty measurement tools may be developed to assess the age-specific standard of living. Policymakers may consider deprivation, in addition to income, as well as change in poverty status, in healthcare and social welfare policymaking to better address the mental health of the population and health inequality in the region.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Longitudinal design with three waves of data collection enabled assessment of change in poverty status over a 7-year period.

• Poverty was assessed using both a validated Deprivation Index, covering material and social aspects of poverty, in addition to conventional income thresholds.

• Recruitment of adults across a wide age range permitted subgroup analyses of the poverty-mental health association by both age and gender.

• High participant attrition over the 7-year follow-up period may introduce selection bias and limit the generalisability of the findings.

Introduction

According to the WHO,¹ one in every eight people (12.5%) were living with a mental disorder worldwide in 2019. Across both year and gender, the two most common mental disorders were depressive and anxiety disorders,^{1 2} and they remained among the leading causes of global burden of disability.² The high prevalence of mental illness not only impairs health in their own right but also increases the risk of other health outcomes,² including increased morbidity and mortality associated with a range of physical conditions.³

In Hong Kong, a special administrative region of China, the estimated prevalence of common mental disorders was 13.3% in the general populations.⁴ Similar to global trend, depressive and anxiety disorders are the most frequent diagnoses locally.

Among people with mental illness burden, it was generally reported that the socioeconomically disadvantaged populations were affected disproportionately. A systematic review reported that people with the lowest incomes in a community suffer 1.5–3 times more frequently from depression, anxiety and other common mental illnesses than those with the highest incomes.⁵ Many other studies also found that the rate of depression, anxiety, substance use disorders and suicide correlate negatively with income.^5–8

In current literature, poverty as measured by income is commonly adopted. Income, however, is only an indirect proxy of poverty as it only measures the monetary aspect of life that contribute to poverty. On the other hand, deprivation, which covers non-monetary and social aspects of poverty, is considered a more comprehensive measurement of poverty than income as it measures lifetime resources and current living standard directly.^{9 10} A longitudinal study found that deprivation, but not income-poverty, was significantly associated with anxiety and stress in Hong Kong.¹¹ Another Japanese study also showed that relative deprivation has a stronger association with depressive symptoms than relative income-poverty among older adults.¹² Therefore, we used the concept of relative deprivation proposed by Townsend in 1979,¹³ referring to individuals who lack the materials or resources that are customary or widely approved in the societies to which they belong, to measure poverty in this study.

Previous evidence revealed that the relationship between socioeconomic status and mental health is bidirectional, when poverty is both a cause and consequence of mental health problems.^{14 15} Nevertheless, many current studies are cross-sectional in nature, which may underestimate the relationship between poverty and mental health due to the possibility of biases from confounders and health selection,¹⁶ and limit the possibilities for drawing causal inferences.¹⁷

While previous cross-sectional studies have demonstrated associations between deprivation and mental health, longitudinal evidence remains sparse, particularly regarding the cumulative or changing nature of poverty over extended periods. Furthermore, although our earlier analysis of the first two waves of this cohort (spanning 2 years) identified associations between deprivation and psychological distress, it remains unclear whether these relationships would persist over a longer time horizon and whether they would differ meaningfully by age or gender. Additionally, among diverse mental health outcomes, anxiety and stress are often more immediate responses to economic hardship. Therefore, these two mental health perspectives are particularly suitable for examination with cross-sectional and shorter-term longitudinal analyses.

To address these gaps, this study extends follow-up to 7 years and examines: (1) the cross-sectional and longitudinal associations of poverty, measured by deprivation and income status, with anxiety and stress among adults in Hong Kong; (2) the impact of changes in poverty status over time on subsequent mental health and (3) potential age- and gender-specific differences in these associations among Hong Kong adults.

Methodology

Study design

This is a longitudinal cohort study. In 2014–2015, we conducted a territory-wide baseline survey (T1) among adults in Hong Kong and recruited 2282 household respondents by stratified random sampling. Then we carried out the first follow-up survey (T2) of this cohort in 2016–2017, where 1476 participants were successfully recruited. This study, as the second follow-up survey (T3), was conducted from December 2020 to September 2022.

Participants

Participants in this T3 study are adults, who were aged 18 years or above, living in Hong Kong. They were not institutionalised and were able to speak Cantonese Chinese. They had participated in either the T1 baseline survey alone or both the T1 baseline and T2 follow-up survey.

Data collection

We originally intended to carry out face-to-face interviews by trained interviewers. However, this was no longer feasible due to the severe outbreaks of COVID-19 in Hong Kong since mid-2021. We then changed the data collection format to phone interview and online questionnaire, by providing quick-response code to participants, using the same standardised structured questionnaire.

All participants gave voluntary informed consent to participate in this study. Written consents were collected during face-to-face interviews and on online questionnaires. Verbal consents were collected during phone interviews.

Measurements

Information regarding participants’ sociodemographic factors, lifestyle factors, poverty and health status was collected.

Sociodemographic factors

We collected information on the sociodemographic characteristics of participants, including age, gender, marital status, education level and occupation. Marital status was further categorised as ‘unmarried’ (including single, divorced, separated or widowed), and ‘married/cohabited’. Education level was further categorised as ‘primary or below’, ‘secondary’ and ‘tertiary or above’. Occupation of participants’ current or last job was further classified as ‘skill level 1’, which included elementary occupations and others, ‘skill level 2’, which included clerical support workers, service and sales workers, craft and related workers, plant and machine operators and assemblers, and ‘skill level 3 or 4’, which included managers, administrators, professionals and associate professionals, based on the assumed required skill levels as suggested by the International Labour Organization.¹⁸ Additionally, students, retired persons and persons who looked after family were also included in the occupation categories as these reflected participants’ economic activity.

Lifestyle factors

We examined the smoking and alcohol drinking habits, as well as physical activity levels of participants. Smoking was classified as ‘non-smoker’ and ‘people who smoke currenly or in the past’. Alcohol drinking habit was assessed by the Alcohol Use Disorders Identification Test-Consumption (AUDIT-C),¹⁹ which is a tool derived from the first three questions of the AUDIT instrument. Participants who scored at least 5 out of 12 items on AUDIT-C were considered potentially ‘risky drinkers’,²⁰ while others were classified as ‘non-risky drinkers’. Physical activity level was assessed using the International Physical Activity Questionnaire (IPAQ) short form.²¹ Participants were categorised as ‘inactive’, ‘minimally active’ and ‘active’ based on the intensity and duration of their physical activity per week according to the IPAQ scoring system.

Poverty measures

In this study, we used equivalised monthly household income, which is derived from dividing the monthly household income by the square root of the total number of persons living in the same household, to measure income-poverty. It is a relative poverty concept as it allows for comparison of households of different sizes.²² In the T1 baseline survey, people having less than half of the median of the equivalised monthly household income, that is, HKD 6059.2 (corresponded to US$ 776.8 approximately), were considered as ‘income-poor’. In this T3 study, we adjusted the income-poverty cut-off with reference to the inflation rate and poverty line in Hong Kong in 2022, and classified people having equivalised monthly household income less than HKD 7028.7 (corresponded to US$ 901.1 approximately) as ‘income-poor’.

In addition to income, we assessed participant’s deprivation status to measure poverty. Participant’s deprivation status was assessed by the Deprivation Index (DI), which is a tool developed by our research team. In the baseline survey, we randomly selected 301 participants and asked if they consider a list of items as necessities in Hong Kong. Eventually, there are 21 items of which more than half of our participants considered as necessities, and these items were included in the construction of DI. Among these 21 items, 17 items are measures of material deprivation, while the remaining 4 items are measures of social deprivation. The resultant DI has been tested on its face validity, construct validity, reliability and additivity and external validity. This DI has high reliability with a high Cronbach’s alpha at 0.832.²³ A detailed description of the development and validation of the DI has been published previously.²³ Considering the weighted mean DI score of the 10 deciles of equivalised monthly household income, deprivation score was particularly higher at 2.66 in the lowest income decile and it dropped significantly in the second and third deciles to 1.55 and 1.32, respectively. Therefore, we considered people who lacked at least two items on the DI due to lack of affordability but not personal preference as ‘deprived’ in this study. A list of items included in the construction of DI is listed in table 1.

Table 1. List of items on Deprivation Index (DI).

1	Three meals a day
2	Fresh fruit or vegetables every day
3	Eat fresh/frozen poultry for special occasions (eg, Chinese New Year)
4	One or two pieces of new clothes in a year
5	Enough warm clothes for cold weather
6	One set of decent clothes (eg, for job interview/Chinese New Year celebration)
7	Able to consult private doctor when you are sick
8	Able to consult Chinese medicine practitioner when you are sick and purchase prescribed medicine
9	Can pay for spectacles if needed
10	Have toilet inside a self-contained apartment, with no need to share with other residents
11	A mobile phone or telephone landline
12	A washing machine
13	An air-conditioner
14	A computer device with internet connection at home
15	Able to replace worn out furniture
16	Able to replace/repair broken electrical goods (eg, refrigerator or washing machine)
17	Some amount of money to spend each week on yourself, not on your family
18	Celebrations on special occasions (eg, Chinese New Year)
19	A meal out with friends or family at least once a month
20	Can offer a gift of money on occasion of wedding
21	Give red pocket money during Chinese New Year

For change in deprivation status throughout T1, T2 and T3, we categorised participants who had participated in all three waves of survey and experienced transition of deprivation status into six groups as shown in table 2. Similar classification approach was applied to participants experiencing transition trajectory in income-poverty status. This categorisation of transistion trajectory in poverty measures was based on previous findings that time dimension, transient and permanent impact, of poverty on health mattered.²⁴

Table 2. Categorisation of participants experiencing change in deprivation status throughout T1, T2 and T3.

	Deprivation status in
	T1	T2	T3
Persistent deprivation	Deprived	Deprived	Deprived
Increased deprivation	Non-deprived	Deprived	Deprived
	Non-deprived	Non-deprived	Deprived
Occasional non-deprivation	Deprived	Non-deprived	Deprived
Occasional deprivation	Non-deprived	Deprived	Non-deprived
Decreased deprivation	Deprived	Deprived	Non-deprived
	Deprived	Non-deprived	Non-deprived
Persistent non-deprivation	Non-deprived	Non-deprived	Non-deprived

Anxiety and stress measurement

Participant’s anxiety and stress symptoms were assessed by the Depression, Anxiety and Stress Scale-21 items (DASS-21) in Chinese version.²⁵ This tool has been validated as a reliable and sensitive self-administered psychological instrument to detect negative emotional syndromes of depression, anxiety and stress^{25 26} and can be applied to Chinese adults.²⁷ Anxiety and stress domains were included in this study. The anxiety domain assesses autonomic arousal, situational anxiety and subjective experience of anxious affect. The stress domain assesses difficulty relaxing, nervous arousal and being easily upset or agitated. Each domain contains seven items scored on a 4-point Likert scale from 0 to 3. Participants were asked to indicate the presence of these symptom(s) over the past week, with 0 indicating not applicable at all and three indicating applicable most of the time. Scores from all items in each domain were summed up and multiplied by two to calculate the final score in each subscale. A higher final score indicates greater severity of the symptoms. Based on the DASS manual,²⁸ we classified participants into five categories according to their corresponding anxiety and stress scores, with the first category indicating normal and fifth category indicating extremely severe form of anxiety or stress. The five categories are ‘normal’, ‘mild’, ‘moderate’, ‘severe’ and ‘extremely severe’, which correspond to score of 0–7, 8–9, 10–14, 15–19 and more than 19 respectively on the anxiety subscale and score of 0–14, 15–18, 19–25, 26–33 and more than 33 respectively on the stress subscale. In data analysis, we used the first category in each subscale, that is, ‘normal’, as the reference group.

Statistical analysis

Categorical variables are presented as numbers and percentages. Multiple linear regression and multivariable ordinal logistic regression were performed to investigate the impact of poverty measures on anxiety and stress score and risk respectively cross-sectionally and longitudinally. Generalised estimating equations were applied to examine the longitudinal association of change in DI and income, and anxiety and stress score across time. We progressively adjusted for variables of sociodemographic and lifestyle factors. Sociodemographic factors were adjusted to estimate the association of poverty measures with mental health outcomes independent of age, gender, socioeconomic position and other sociodemographic effects. Lifestyle factors were adjusted because they may confound or mediate the poverty-mental health pathway, and we aimed to estimate the direct associations. We used the most up-to-date information of covariates collected in T3 for confounder adjustments. Deprivation and income-poverty status are also mutually adjusted in all models. Multicollinearity between these two measures was assessed using variance inflation factors (VIF). VIF for deprivation was 1.124 and for income-poverty was 1.515, both well below the conventional threshold of 10 (or even the more conservative threshold of 5), indicating little to no concern for multicollinearity. For longitudinal association, the scores and ORs are further adjusted for baseline corresponding anxiety and stress score.

For the sensitivity analyses, attrition weighting was applied to account for loss to follow-up. Attrition weights were estimated using logistic regression with baseline measures, including sociodemographic characteristics, poverty measures and stress and anxiety scores. Weighted estimates were used to assess the robustness of the cross-sectional associations of T3 poverty measures and mental health outcomes, as well as the longitudinal associations of baseline poverty measures and T3 mental health outcomes.

Subgroup analyses in terms of gender, including male and female subgroups, and age, including 18- to 59-year-old and 60-year-old or above subgroups, were performed.

We performed statistical analysis using SPSS V.29. All statistical tests were two-tailed with a significance level of p-value less than 0.05.

Patient and public involvement

Patients were not involved in the design and conduct of this research. Public was involved in the development of the DI and the conduct of this research.

Results

In this study, we successfully recruited 508 participants for follow-up. The recruitment process is illustrated in figure 1. Nevertheless, 15 of them had unmatched gender and age profile compared with the baseline survey, likely being another family member living in the same household, so they were ultimately excluded from data analysis to ensure the quality of data. Eventually, 493 participants remained for data analysis. Characteristics of participants are shown in table 3.

Figure 1. Among the 2282 T1 participants, 1807 of them provided phone number for contact. Then they were contacted in this T3 study. When 344 phone numbers were no longer valid, 1463 phone numbers were remained for further contact. With 955 refusals and no one answering phone call, 508 participants were successfully enumerated in T3 eventually.

Table 3. Characteristics of participants (n=493).

Sociodemographic characteristics		N	%
Gender
	Male	177	35.9
	Female	316	64.1
Age
	18–30	44	8.9
	31–40	48	9.7
	41–50	110	22.3
	51–60	91	18.5
	61–70	93	18.9
	≥71	107	21.7
Marital status
	Single/divorced/separated/widowed	175	35.5
	Married/cohabited	318	64.5
Education level
	Primary or below	132	26.8
	Secondary	247	50.1
	Tertiary or above	114	23.1
Occupation
	Skill level 3 or 4	66	13.4
	Skill level 2	125	25.4
	Skill level 1	61	12.4
	Student	6	1.2
	Looked after family	64	13
	Retired	170	34.5
	Unable to work due to disability/illness	1	0.2
Lifestyle factors
Smoking status#a
	Non-smoker	421	85.7
	Current/past smoker	70	14.2
Alcohol drinking#b
	Non-risky drinker	471	97.3
	Risky drinker	13	2.7
Physical activities
	Inactive	106	21.5
	Minimally active	280	56.8
	Active	107	21.7
Poverty measures
Deprivation status
	Non-deprived	337	68.4
	Deprived	156	31.6
Income-poverty status
	Non-income-poor	339	68.8
	Income-poor	154	31.2

Missing data #a:2, #b:9.

Baseline characteristics of T3 participants and drop-outs are also compared and shown in table 4. It shows that T3 participants tended to be significantly younger, women and highly educated, but with no significant difference in terms of poverty status, anxiety and stress status.

Table 4. Comparison of baseline characteristics of T3 participants and drop-outs.

Baseline (T1) characteristics	T3 participants	Participants who dropped out in T3	Pearson χ2 p-value
Gender
Female	63.5%	58.1%	0.031
Age (mean±SD, years)	49.8±16.6	52.7±18.2	<0.001
Marital status
Married/cohabited	66.1%	61.9%	0.093
Education level			<0.001
Primary or below	26.0%	35.4%
Secondary	55.8%	51.2%
Tertiary or above	18.3%	13.4%
Occupation (current or last jobs)			0.131
Skill level 3 or 4	14.7%	11.0%
Skill level 2	34.3%	38.0%
Skill level 1	21.1%	23.0%
Student	5.8%	4.9%
Looked after family/home	24.2%	23.2%
Smoking status
Current/past smoker	14.0%	19.2%	0.008
Alcohol drinking
Risky drinker	2.3%	3.3%	0.259
Physical activities			0.087
Inactive	71.6%	76.1%
Minimally active	17.4%	13.8%
Active	11.0%	10.2%
Deprivation status
Deprived	18.1%	19.6%	0.488
Income-poverty status
Income-poor	18.6%	16.6%	0.304
Anxiety level			0.265
Normal	88.2%	90.7%
Mild	5.9%	3.6%
Moderate	2.6%	2.4%
Severe	1.4%	1.4%
Extremely severe	1.8%	1.9%
Stress level			0.341
Normal	94.5%	94.6%
Mild	2.2%	1.7%
Moderate	1.2%	2.3%
Severe	1.6%	0.9%
Extremely severe	0.4%	0.5%

Bold typeface indicates statistical significance.

Among these T3 participants, 16 of them (3.2%) completed the face-to-face interviews, 464 (94.1%) finished the phone interviews and 13 (2.6%) filled in the online questionnaire. On the other hand, data collection period for the T3 study was from December 2020 to September 2022; that is, before and during the severe local outbreak of COVID-19. A sensitivity analysis is carried out by dividing the dataset based on the data collection cut-off date when mass vaccination was implemented in Hong Kong (ie, on 26 February 2021). It revealed no significant difference in anxiety or stress scores between samples interviewed before and after this cut-off date.

Table 5 shows the cross-sectional association of T3 poverty measures with T3 anxiety and stress. In the multiple linear regression, deprivation was statistically significantly associated with higher anxiety score (unstandardised beta-coefficient 1.41, 95% CI 0.59 to 2.24) and stress score (unstandardised beta-coefficient 1.77, 95% CI 0.59 to 2.94) than non-deprivation in T3. In the multivariable ordinal logistic regression, deprived participants were also significantly associated with higher risk of more severe forms of anxiety (OR 1.81, 95% CI 0.93 to 3.53) and stress (OR 2.07, 95% CI 0.89 to 4.78) than non-deprived participants in T3, though the adjusted ORs were not statistically significant. However, income-poverty did not demonstrate any significant association with anxiety and stress score and risk cross-sectionally.

Table 5. Cross-sectional and longitudinal association of poverty measures with T3 anxiety and stress.

		T3 anxiety score		T3 anxiety risk		T3 stress score		T3 stress risk
	n (%)	Unstandardised B coefficient (95% CI)	P value	Adjusted OR (95% CI)	P value	Unstandardised B coefficient (95% CI)	P value	Adjusted OR (95% CI)	P value
Cross-sectional association
Deprivation status in T3
Non-deprived	337 (68.4)	Reference		Reference		Reference		Reference
Deprived	156 (31.6)	1.41 (0.59 to 2.24)	<0.001	1.81 (0.93 to 3.53)	0.082	1.77 (0.59 to 2.94)	0.003	2.07 (0.89 to 4.78)	0.090
Income status in T3
Non-income-poor	339 (68.8)	Reference		Reference		Reference		Reference
Income-poor	154 (31.2)	0.56 (−0.40 to 1.52)	0.251	1.18 (0.54 to 2.59)	0.677	0.51 (−0.86 to 1.87)	0.466	1.47 (0.54 to 4.04)	0.451
Longitudinal association*
Deprivation status in T1
Non-deprived	353 (81.9)	Reference		Reference		Reference		Reference
Deprived	78 (18.1)	−0.38 (−1.43 to 0.68)	0.482	1.11 (0.40–3.05)	0.843	0.10 (−1.42 to 1.63)	0.893	0.91 (0.28–3.02)	0.880
Income status in T1
Non-income-poor	393 (83.1)	reference		reference		reference		reference
Income-poor	80 (16.9)	−0.10 (−1.20 to 1.01)	0.867	0.70 (0.21–2.27)	0.549	−0.27 (−1.87 to 1.33)	0.740	0.79 (0.20–3.11)	0.736

Anxiety and stress risk: first to fifth category, first category (normal) as reference group.

Adjusted for age, gender, marital status, education level, occupation, smoking status, alcohol drinking and physical activities.

ORs of all listed variables were mutually adjusted.

Bold typeface indicates statistical significance.

^*Further adjusted for T1 baseline corresponding anxiety/stress scores.

The longitudinal association of baseline poverty measures with T3 anxiety and stress is presented in table 5. Baseline deprivation and income status were not statistically associated with T3 anxiety and stress outcomes. Total number of participants under longitudinal analysis for deprivation status was 431 while that for income status was 473. The number of participants included in the deprivation status analysis is smaller than that in the income status analysis because some participants’ T1 deprivation assessment results were used to develop the DI.

Table 6 shows the association of change in poverty measures throughout T1–to T3, and anxiety and stress in T3. Participants experiencing increased deprivation since baseline tended to have higher anxiety score (unstandardised beta-coefficient 1.01, 95% CI 0.05 to 1.97) and stress score (unstandardised beta-coefficient 1.68, 95% CI 0.30 to 3.07) in T3 than participants with persistent non-deprivation. Meanwhile, decreased deprivation was significantly associated with lower anxiety score in T3 (unstandardised beta-coefficient −1.41, 95% CI −2.78 to −0.03) than persistent non-deprivation. For change in income status throughout T1–T3, participants who were income-poor in both T1 and T3, but became non-income-poor in T2, tended to have significantly higher anxiety score (unstandardised beta-coefficient 4.33, 95% CI 1.78 to 6.87), higher risk of severe form of anxiety (OR 16.99, 95% CI 2.57 to 112.35) and stress (OR 10.33, 95% CI 1.31 to 81.81) in T3 than participants with persistent non-income-poverty.

Table 6. Association of change in poverty measures throughout T1, T2 and T3, and anxiety and stress in T3.

		T3 anxiety score		T3 anxiety risk		T3 stress score		T3 stress risk
	n (%)	Unstandardised B coefficient (95% CI)	P value	Adjusted OR (95% CI)	P value	Unstandardised B coefficient (95% CI)	P value	Adjusted OR (95% CI)	P value
Change in deprivation status throughout T1–T3
Persistent deprivation	15 (4.2)	0.86 (−1.30 to 3.01)	0.435	2.48 (0.38 to 16.34)	0.347	2.22 (−0.89 to 5.33)	0.162	5.23 (0.70 to 38.92)	0.106
Increased deprivation	90 (25.0)	1.01 (0.05 to 1.97)	0.040	1.28 (0.43 to 3.82)	0.664	1.68 (0.30 to 3.07)	0.017	1.86 (0.57 to 6.04)	0.301
Occasional non-deprivation	21 (5.8)	−0.69 (−2.51 to 1.14)	0.460	0.86 (0.11 to 6.86)	0.883	−0.81 (−3.46 to 1.85)	0.550	0.60 (0.04 to 10.35)	0.725
Occasional deprivation	7 (1.9)	−0.18 (−3.30 to 2.94)	0.910	7.20 (0.43 to 120.08)	0.169	−0.56 (−5.06 to 3.94)	0.807	7.27 (0.33 to 158.61)	0.207
Decreased deprivation	36 (10.0)	−1.41 (−2.78 to −0.03)	0.045	1.29 (0.27 to 6.18)	0.747	−0.38 (−2.36 to 1.60)	0.707	1.71 (0.27 to 11.01)	0.573
Persistent non-deprivation	191 (53.1)	Reference		Reference		Reference		Reference
Change in income status throughout T1–T3
Persistent income-poverty	39 (10.0)	0.50 (−1.00 to 1.99)	0.513	0.61 (0.05 to 7.20)	0.692	0.47 (−1.70 to 2.63)	0.673	<0.001 (0)	0.999
Increased income-poverty	87 (22.3)	0.36 (−0.70 to 1.41)	0.509	2.38 (0.62 to 9.16)	0.208	0.33 (−1.20 to 1.85)	0.676	2.30 (0.49 to 10.80)	0.293
Occasional non-income-poverty	10 (2.6)	4.33 (1.78 to 6.87)	<0.001	16.99 (2.57 to 112.35)	0.003	3.64 (−0.03 to 7.31)	0.052	10.33 (1.31 to 81.81)	0.027
Occasional income-poverty	6 (1.7)	−0.37 (−3.74 to 3.00)	0.830	0.76 (0.05 to 12.55)	0.850	0.00 (−4.83 to 4.84)	0.999	0.69 (0.03 to 14.91)	0.812
Decreased income-poverty	23 (6.4)	−0.70 (−2.42 to 1.02)	0.426	<0.001 (0)	0.999	−0.41 (−2.89 to 2.07)	0.743	<0.001 (0)	0.999
Persistent non-income-poverty	225 (62.5)	Reference		Reference		Reference		Reference

Anxiety and stress risk: first to fifth category, first category (normal) as reference group.

Adjusted for age, gender, marital status, education level, occupation, smoking status, alcohol drinking, physical activities and T1 baseline corresponding anxiety/stress scores.

ORs of all listed variables were mutually adjusted.

Bold typeface indicates statistical significance.

Regarding increased deprivation over time, table 7 further shows the results of generalised estimating equations that for each unit increase in DI throughout T1, T2 and T3, participants had a statistically significant increase in anxiety and stress scores of 0.31 and 0.36, respectively, across time. On the other hand, for each 500 HKD increase in equivalised monthly household income throughout T1–T3, participants experienced slight decrease in anxiety and stress score with only marginal statistical significance.

Table 7. The association between change in poverty measures and change in anxiety and stress scores throughout T1, T2 and T3.

	Anxiety score		Stress score
	Unstandardised B coefficient (95% CI)	P value	Unstandardised B coefficient (95% CI)	P value
Change in Deprivation Index (DI) throughout T1, T2 and T3
1 unit increase in DI	0.31 (0.18 to 0.45)	<0.001	0.36 (0.18 to 0.55)	<0.001
Change in income throughout T1, T2 and T3
500 unit increase in income	−0.01 (−0.02 to 0.007)	0.068	−0.02 (−0.03 to 0.001)	0.073

Bold typeface indicates statistical significance.

Adjusted for three time points (as continuous variables), age, gender, marital status, education level, occupation, smoking status, alcohol drinking and physical activities.

In order to understand the demographic profile difference in depth, subgroup analyses, in terms of age and gender, were also performed. It was found that younger adults, aged less than 60 years, who were income-poor in T3 tended to have significantly higher anxiety and stress scores in T3 than their non-income-poor counterparts in the same age group. Furthermore, younger adults with occasional non-income-poverty throughout T1–T3 also tended to have higher anxiety and stress scores in T3 than persistently non-income-poor younger adults. However, older adults, aged 60 years or above, did not demonstrate these cross-sectional and longitudinal associations between income-poverty, and anxiety and stress score. On the other hand, deprived women, but not men, generally tended to show higher anxiety and stress scores than non-deprived women cross-sectionally and longitudinally.

Discussion

Consistent with our previous baseline cross-sectional analysis,¹¹ our T3 cross-sectional findings showed that individual-level deprivation was significantly associated with higher anxiety and stress scores, even after being adjusted for income and other covariates. A 1.41-point higher anxiety score on the DASS-21 anxiety subscale corresponds to approximately 0.18 SD in our sample, while a 1.77-point higher stress score on the stress subscale represents approximately 0.19 SD. According to conventional benchmarks (Cohen’s d), these are small effects. But more importantly, our ordinal logistic regression showed that deprived participants had 1.81 times higher odds of being in a more severe anxiety category, and 2.07 times higher odds for stress, compared with non-deprived participants. To illustrate the absolute difference: among non-deprived participants, approximately 2.1% had ‘severe’ or ‘extremely severe’ anxiety, whereas among deprived participants this proportion was 5.4% (using our sample margins; a difference of 3.3% points). This means that for every 100 deprived individuals, about three additional people experience clinically severe anxiety that would warrant clinical attention per DASS-21 guidelines. In contrast, income-poverty showed no significant association with anxiety and stress scores. This pattern supports the view that deprivation, a direct measure of affordability and living standard, captures dimensions of poverty relevant to mental health that income alone does not. The non-significant findings for anxiety and stress severity categories likely reflect that score increases were not large enough to shift participants into a higher severity classification. Nevertheless, small differences in anxiety and stress scores can be clinically meaningful when they affect daily functioning or predict downstream outcomes, particularly when they accumulate at the population level to a meaningful burden.

Also, our results suggested that neither baseline deprivation nor income-poverty could predict anxiety or stress outcomes in T3. This may be due to the long duration between baseline and second follow-up survey that spanned 7 years, during which people might have experienced change in poverty status that could result in different mental health outcomes eventually. Despite this non-significant longitudinal relationship of one-off poverty measure and subsequent mental health outcomes, we found statistically significant longitudinal association between change in poverty measures throughout 7 years and subsequent mental health outcomes.

Longitudinally, as compared with persistent non-deprivation, increased deprivation over the period from T1 to T3 was significantly associated with higher T3 anxiety and stress scores. This is consistent with our previous T1 and T2 findings.¹¹ Each additional item lacking on the DI across waves corresponded to significantly elevated scores. This pattern aligns with Relative Deprivation Theory, which postulated that worsening circumstances relative to one’s past or peers may engender frustration and psychological distress.^{29 30} In contrast, persistent deprivation did not reach statistical significance, unlike in our previous T1 and T2 findings.¹¹ While point estimates remained elevated for this group, the non-significant finding may reflect either true adaptation to chronic hardship over a longer time horizon or limited statistical power due to the small number of persistently deprived participants in the T3 sample. This plausible explanation of adaptation warrants further empirical studies with larger sample size in the future.

Also, participants with occasional non-income-poverty (ie, those who escaped income-poverty at T2 but finally relapsed by T3) exhibited significantly higher risk of severe anxiety and stress than the persistently non-income-poor. This ‘yo-yo’ pattern may be particularly distressing, as temporary improvement followed by renewed hardship could heighten frustration, disappointment and perceived loss. This speculated explanation and interpretation aligns well with evidence that income losses exert stronger negative effects on well-being than equivalent gains, as shown in studies from Sweden³¹ and Germany and the United Kingdom,³² and that unmet expectations contribute to frustration and stress.³³

Interestingly, subgroup analyses showed that younger adults who had income-poverty in T3 and occasional non-income-poverty over time throughout T1–T3 tended to have significantly higher anxiety and stress scores in T3 than their non-income-poor counterparts. However, older adults did not demonstrate this association between income-poverty and mental health outcomes. This finding aligned with the conclusion from a systematic review that income changes have a causal effect on the mental health of working-age adults, aged 16–64 years.³⁴ Additionally, another longitudinal study showed that lower income level was significantly associated with higher likelihood of mental disorders among people aged 20- to 54 years old but this association was not significant among people aged 55 years or older in the USA.⁵ Generally, younger adults have less assets and more dependents than older adults or retirees. They are usually not eligible for governmental or organisational subsidy or welfare programme in Hong Kong and rely heavily on income alone to make their ends meet. Therefore, when they encountered income-poverty, especially long-term one, they experienced much frustration and anxiety. Santiago et al³⁵ proposed that lower income levels restrict financial resources to provide basic necessities and thus create tremendous financial stress, which contributes to anxiety. Therefore, younger population may be more sensitive and vulnerable to financial constraints and have stronger emotional reaction towards income-poverty than their older counterparts. On the other hand, some studies suggested that older persons may experience more multidimensional poverty than the younger ones in Taiwan³⁶ and South Korea.³⁷ Therefore, deprivation assessment, covering non-monetary and social aspects of poverty, may be a more appropriate tool than income alone to assess poverty situation, particularly among older persons.

Furthermore, deprived women, but not men, generally tended to have higher anxiety and stress scores than non-deprived women cross-sectionally and longitudinally in this study. Current literature showed that economic disadvantage and poverty, as well as gender inequities and discrimination, being the risk factors for the gender difference in mental health disorder prevalence and symptoms.^{38 39} It is possible that women are more likely to express their worries and fears. Conversely, men are more likely to contain their fear and insecurity.⁴⁰ Another possibility could be the relatively larger sample size of women (64.1%) than men (35.9%) in this T3 study.

Additionally, this study applied multiple linear regression, multivariable ordinal logistic regression and generalised estimating equations to analyse the associations of deprivation and income with anxiety and stress from different dimensions. There may be potential inflation risk of type I error with distinct statistical models. In this study, all statistical tests were two-tailed with a significance level of p-value set at less than 0.05. If the Bonferroni correction is applied, the new p-value threshold could be considered as 0.017 (0.05/3). Considering the statistically significant findings in the cross-sectional analyses in table 5, all the p-values are lower than 0.003, significantly lower than this new p-value threshold. On the other hand, considering the statistically significant findings in the trajectory analyses in table 6, some of the p-values are still lower than 0.017, though some others are higher than this new p-value threshold. In summary, while the potential risk of type I error cannot be completely ruled out, the overall error rate remains limited and controlled in this study.

Limitations

This study has some limitations. First, sample attrition was substantial, with a retention rate of 21.6% over the 7-year follow-up period, partly due to difficulties in re-contacting participants during the COVID-19 pandemic, as well as death and migration. Nevertheless, we tried to recruit and interview participants using diverse modes, including in-person and phone interview as well as online questionnaire, and increased the incentive for each successfully enumerated participant. There is no significant difference in the sociodemographic characteristics, poverty status, anxiety and stress status among participants in the three modes of data collection. Sensitivity analyses using attrition weighting were also carried out and the results were consistent with the main analysis, with all previously significant associations remaining significant. Several associations that were not statistically significant in the unweighted analysis became significant after weighting, including the cross-sectional association of T3 deprivation and anxiety and stress risks; cross-sectional associations of T3 income-poverty with anxiety and stress scores and risks; and the longitudinal association between T1 deprivation and T3 anxiety score. On the other hand, comparison of T3 participants with those lost to follow-up revealed that T3 participants tended to be significantly younger, women and highly educated, but no significant difference in terms of poverty status, anxiety and stress status as shown in table 4. While this pattern suggests that attrition may be unrelated to the key variables of interest given observed covariates, we cannot fully exclude the possibility of selection bias as well as mode effects on reporting. Findings should therefore be interpreted with appropriate caution.

Second, the number of participants in some subgroups, especially the age and gender subgroups, tended to be relatively small due to limited overall sample size. Therefore, this study just presented the preliminary findings of age and gender subgroup analyses. Future study with larger sample size is warranted. Additionally, considering the potential impact of the pandemic on our samples, we carried out a sensitivity analysis to compare the T3 anxiety and stress scores between participants interviewed before and after the mass vaccination implementation in Hong Kong. We found no significant difference in their scores; thus, the potential impact of the pandemic on our samples should be limited.

Third, all questions are self-reported so recall bias might occur. However, most questions covered participants’ experience in previous 1 week to 1 month so the time lapses between events and data collection should have been minimised. Besides, a subset of sample was enumerated with missing data in lifestyle factors. This may introduce bias and reduce power of the study. However, single imputation technique of last observation carried forward was applied to control for potential non-response bias, increase the precision of estimates and thus improve the validity of our results.

Additionally, single DI was applied to assess the deprivation status of all participants in this study. Nevertheless, there may be disagreement regarding which items reflect basic necessities between younger and older adults⁴¹ as their needs could be quite different. Therefore, it may be necessary to develop and test age-specific deprivation assessment tools for different age groups to best assess their age-specific needs and living standard in the future.

Moreover, this study prioritised investigating the associations of poverty with anxiety and stress using DASS-21 because they are often more immediate responses to economic hardship and thus particularly suitable for the cross-sectional and shorter-term longitudinal analyses in this study. Although depression subscale in DASS-21 is not examined in this study, future longitudinal studies with longer follow-up period are relevant to examine this mental health outcome specifically.

Furthermore, although our longitudinal design and adjustment for baseline mental health scores help establish temporal precedence, the observational nature of the study precludes definitive causal inference, and reverse causality, whereby poor mental health leads to subsequent poverty, cannot be fully excluded.

Lastly, as this study was carried out in Hong Kong, the findings and implications may be more relevant to developed and rapidly developing Asian economies with comparable sociocultural context (e.g. Singapore, South Korea, urban cities in China) but may not generalise to rural areas, lower-income countries, or settings with different social welfare structures. Cross-national comparative studies would be valuable to assess the consistency of the deprivation-mental health association across diverse contexts.

Conclusions

Individual deprivation at follow-up and increased deprivation over time from baseline tended to be associated with higher follow-up anxiety and stress scores cross-sectionally and longitudinally. Income-poverty, however, did not demonstrate significant association with anxiety and stress generally. Younger adults may be more affected by income-poverty cross-sectionally and over time, in terms of anxiety and stress score, than older adults. It is important for policymakers to consider deprivation, in addition to income, as well as change in poverty status, as complementary strategy in assessing poverty in healthcare and social welfare policymaking to improve the mental health of the population and address health inequality in the region. We also suggest developing and testing age-specific poverty measurement tools for assessing age-specific needs and standard of living in future research.

Data availability statement

Data are available upon reasonable request.