Study design
Kolkata metropolis in India is home to 4.6 million people, with the fastest rising aging population in any Indian city, making it suitable for our study38. The Kolkata Municipal Corporation provides lists of apartment complexes around the city and 28 apartment complexes were selected based on their representative demographic and socio-economic structure. A random sampling methodology was used to recruit participants. In the first stage, a pre-survey questionnaire and documentation describing the purpose and the intended outcome of the study along with any risks of taking part in the survey was shared with potential participants and informed consent (consent form I) was received from all participants before disbursement of the self-reporting survey questionnaire to a select group (see Study Population). Next, the select group received the survey questionnaire along with a second set of informed consent form informing the participants that their responses at this stage would be used in the final analysis (consent form II). We used a cross-sectional design to study the impact of community groups by classifying the apartment complexes into two categories, those with active community groups during the lockdown (target group) and those without (control group). We started our initial pre-survey outreach in June 2022 and concluded our survey in November 2022. The supplementary information contains the details of the data collection process (SI Fig. 1).
Fig. 1
Government Response Stringency Index in India as calculated by the University of Oxford. The Government of India declared the lockdown on 24 March 2020 as indicated by the green circle in the diagram. The index is a composite of nine response indicators regarding closures of schools, offices, stay-at-home requirements, cancellation of public events, closures of public transport, public information campaigns, restrictions on internal and international travel, with 100 being the strictest condition of lockdown. Our data collection period is from the Summer of 2022 to November 2022. The original source of the government response stringency index: Hale, T., Angrist, N., Goldszmidt, R. et al. A global panel database of pandemic policies (Oxford COVID-19 Government Response Tracker). Nat Hum Behav.5, 529–538; https://doi.org/10.1038/s41562-021-01079-8 (2021).
Lockdown in India
On 24 March 2020, Government of India announced a complete nationwide lockdown to limit the spread of the coronavirus. The Oxford University tracker gave the lockdown measures in India a score of 100, the highest stringency score in the world (Fig. 1). While the impact of the lockdown in containing the spread of the SARS-CoV-2 virus is still being investigated, initial reports suggest that without the lockdown, India would have seen 31,000 more coronavirus cases in a month than it saw in reality39. This complete lockdown sparked a high degree of social isolation that makes India an ideal test case for our study.
Study population
To an initial list of 952 adults aged 65 and over residing across these 28 apartment complexes, we first distributed a pre-survey questionnaire soliciting responses about their mental and physical health as well as their social connection, in particular, if they resided with other family members (besides their spouses) in a multi-family household. This questionnaire also contained details of the study including its purpose and intended outcome as well as details of their intended participation and any associated health risks. Responses and signed informed consent (form I) was received from 794 adults (83.4% response rate) who met the criterion that they did not have multifamily connections or any independent social networks during the lockdown (beyond possible community groups) that would preclude them from the study. These adults received the survey questionnaire and another set of informed consent form (form II).
From the set of responses received in this second stage, we narrowed down our final sample based on a set of exclusion criterion. Following previously cited literature, the exclusion criterion for final sample selection included previous mental health condition requiring anxiolytic agents, opioids, and other stimulants as well as ongoing health challenges like cancer, heart disease, stroke, autoimmune disorders or any other physical ailment that required regular medical intervention. Since we were interested in collecting data on the cardiovascular parameters to study the impact of loneliness on physical health, we also asked the volunteers if they tracked their blood pressure and heart rate regularly before and during the lockdown phase. Finally, we excluded participants who did not provide informed consent. With the exclusion criterion, we finalized a sample of 618 adults, with 346 males (55.99%) and 272 females (44.01%). The final data was anonymized before further analysis to remove any individual identifiers.
Outcome variables
Increase in feelings of loneliness during the lockdown
Our fundamental variable of interest was any perceived change in loneliness before and during the Covid-19 lockdown period in India. A modified version of the University of Los Angeles Revised Loneliness Scale was used to collect responses32. The scale was modified to include change responses or Δ (comparing lockdown to pre-lockdown experience) to ten questions (corresponding original UCLA LS3-SF3 question number in parenthesis): (1) During the lockdown, I feel more alone than before the lockdown (#4); (2) Over the lockdown period, I feel I am no longer close to anyone compared to before (#7); (3) During the lockdown, there has been a reduction in the number of people I can talk to than before (#19); (4) I feel more isolated during the lockdown than before (#14); (5) Due to the lockdown, I am no longer a part of a friend’s group that I used to be before (#5—answer n/a if not a part of a group before); (6) Due to the lockdown, I have more trouble finding people I can talk to than before (#19); (7) Due to the lockdown, I have more trouble finding people I can turn to when I need than before (#20); (8) Due to the lockdown, I feel more left out than I felt before (#11); (9) During the lockdown, I have lost people I used to feel close to (additional question not included in UCLA LS3-SF3) and (10) I feel more depressed over the lockdown period than I did that I think is the result of my increased loneliness (additional question not included in the UCLA LS3-SF3). We used the same four-point frequency scale as in UCLA LS3-SF3, with 1 corresponding to “Never or Do not agree at all” and 4 corresponding to “Often or Strongly agree.” A rating of 1–2 was classified as mild or no change in feelings of loneliness, 3 corresponded to moderate increase in feelings of loneliness, while 4 corresponded to substantial increase in feelings of loneliness. The Cronbach’s alpha measure of consistency was 0.91, indicating strong internal consistency40.
Physical and mental health parameters
To measure the physical impact of loneliness during the lockdown, we collected data on cardiovascular parameters: (a) systolic blood pressure, (b) diastolic blood pressure, and (c) heart rate. Additionally, we collected data on forgetfulness, considered as an early symptom of Alzheimer’s disease, found to be closely related to loneliness. We asked the respondents to report: (i) average systolic and diastolic reading before and during the lockdown phase; (ii) average heart rate before and during the lockdown phase; and (iii) responses, on a scale of 1–4, to the question: “Do you feel you have become more forgetful or have a difficult time remembering things since the lockdown as compared to before?”.
Potential confounders
In order to control for factors beyond the lockdown that could potentially explain any variation in the degree of loneliness observed or any deterioration in the physical and memory related parameters, we used population-based previous cohort studies to control for factors associated with loneliness and physical and mental health classified into four categories: evolutionary changes41, neuroscience related42, epidemiological43, clinical44, and developmental45 factors.
Demographic factors
We collected data on sex, body mass index, and ages of the participants. India comprises of 28 states and 8 union territories, each with its unique language and culture, and people from many states reside in the Kolkata metropolis. Therefore, we also gathered information on familiarity with Bengali, the local language of Kolkata. This was to control for the fact that people might have an easier time connecting if they speak the local language. Additionally, we also collected data on the number of years that they were residing in the apartment complex as these factors are critical for social integration.
Socioeconomic factors
We collected data on employment (currently employed in a job, currently running a business, retired) and current involvement in community organizations or social clubs (classified as “volunteering”), income information (includes pension, investment income, rental income, and income from other sources), education level (less than high school, high school, vocational school, junior or technical college, university, graduate school, or other), marital status (married, widowed/divorced, never married), whether the individual was living alone or with a spouse or partner. In case of joint income of the participants, we calculated the average income of each participant by dividing the annualized income of the household by the number of members (18 and over).
Health, lifestyle, and family history information
We collected data on ongoing chronic health issues including blood pressure, cholesterol, diabetes and other age-related chronic ailments. In addition, since lifestyle has been associated with loneliness46, we also collect data on exercise and physical activity of our sample, particularly changes in physical activity during the lockdown period (no change, some change, substantial change) as well as change in sleep patterns (no change, some change, substantial change). In addition, we also collected data on alcohol and tobacco consumption of the participants. Given the evolutionary and genetic factors associated with loneliness, we surveyed the participants to get the family history of mental health challenges by posing the question, “To the best of your knowledge, do you have any immediate family member that was or has been diagnosed or suffering from loneliness or depression? (yes/no/not sure)”.
Statistical analysis
First, we used descriptive statistics (mean/sample proportion and standard deviation) of the potential confounders and their distribution across the three classifications of loneliness determined by the UCLA LS3-SF3 scoring groups to detect patterns in confounders that might explain severity of loneliness, noting any differential impact through ANOVA testing.
Second, we used a random-effects probit regression model to study the differential impact of community groups on the degree of loneliness with the target group coded as 1 and the control group coded as 0. Random effects were used to control for the unobservable sample characteristics. To construct the probit regression model with a binary dependent variable, for each classification, the dependent variable was assigned a value of 1 if the participant belonged to that classification as determined by the survey responses and UCLA LS3-SF3 scoring guidelines, and 0 otherwise. Pseudo R-squared measures were calculated for goodness of fit47. In addition, we conducted a two-tailed test of difference of means and proportions across the target and control samples to make sure that they were comparable and sample heterogeneity beyond the treatment was not driving our results. p-values < 0.05 were considered statistically significant.
Third, we examined the association between the perceived changes in cardiovascular measures and potential risks of Alzheimer’s disease as measured by increased forgetfulness and the degree of loneliness experienced during the Covid-19 lockdown. Given the risk of erroneous self-reported measurements, we used a more conservative estimate of what we considered as an “increase” in these parameters. We defined any observed mean cardiovascular parameter during the lockdown period at least two standard deviations higher than the mean during the pre-Covid period as indicative of worsened cardiovascular health in our baseline (we relaxed the measure to at least one standard deviation in robustness checks), while any change less than one standard deviation was deemed mild or negligible under both baseline and robustness tests. Similarly, a forgetfulness score of 1 or 2 was deemed mild or negligible while 3 or 4 was deemed as substantial, following such categorizations in previous research. The odds ratios (ORs) of the incidences of a rise in systolic pressure, diastolic pressure, heart rate, and indicators of forgetfulness measured across the loneliness classifications were calculated using a multinomial logistic regression model, controlling for potential confounders. We built our models using a “stack” framework where the confounding factor groups were added incrementally such that model 1 includes the demographic factors, model 2 includes the demographic and socioeconomic factors, while model 3 is the complete model including all confounding factors. The p-trends were calculated using a generalized linear model.
Finally, we conducted another set of random-effects probit regressions to examine if having active community groups in apartments help mitigate to some extent the impact of loneliness on physical and mental health. Two alternative methodologies were applied for consistency check. For our main analysis, we used a random-effects probit model with each health measure coded as a binary variable taking a value of 1 in case of a reported increase in the observed measure, and 0 otherwise. The impact of community groups was analyzed separately for each classification of loneliness. In all instances, the complete model including all confounders (model 3) was used for the analysis, with appropriate random effects included to control for any unobserved heterogeneity. The pseudo R-squared measures were reported in each case for goodness of fit. Additionally, as a robustness check, we also conducted another set of probit regressions on the complete sample of 618 participants. Two binary variables and their interactions were used for the purpose, where “Mild or no change in loneliness” takes a value of 1 for incidences of mild or no change in loneliness reported during the lockdown, and 0 for incidences of moderate or substantial increases in loneliness. Similarly, “Community groups” takes a value of 1 for apartments that had active community groups during the lockdown, and 0 otherwise. The dependent variable remained the same in both setups, and appropriate confounders and random effects were included in all instances.
We used Python 3.11 for data cleanup and anonymization and imported the finalized dataset to STATA version 14 (Stata Corp LLC) for our analysis. Any p-value < 0.05 (95% confidence interval in a two-tailed test) was considered as statistically significant for the interpretation of the results. The command “ttest” was used to compare difference in means and the command “prtest” was used to compare any difference in proportional values across the target and control groups for the set of confounding factors controlled for in the regression analysis. The figures were created by the authors using excel software by importing the results from STATA.
Ethics declaration
This study includes survey of human participants. All procedures were conducted following the ethical standards of the Helsinki Declaration of 1975, as revised in 2013. The experimental steps and protocol were reviewed and approved by the ACSEF & B.K. Roy Institutional Review Board as a part of a larger, multi-year study. The purpose, design, any potential risks associated with the survey, and the expected outcome of the study was shared with the participants and informed consent was received. No human images were used in the study and final results were all anonymized and stripped off any identifying and sensitive participant information. No minors were included in the study and no compensation was given to the participants for taking part in the study.
