Study participants

The Japan Gerontological Evaluation Study (JAGES) was established in 2010, sampling residents aged 65 or older nationwide. One of the field sites of the JAGES cohort is based in the Iwanuma City (total population of 44,187 in 2010). We mailed questionnaires to all older residents in August 2010 (n = 8,576), using the city’s residential register. The survey covered personal characteristics, lifestyle and health, with a 59.0% response rate (n = 5,058), which is comparable to other surveys of community-dwelling residents (Santos-Eggimann et al., Reference Santos-Eggimann, Cuénoud, Spagnoli and Junod2009).

The 11 March 2011 earthquake and tsunami severely affected Iwanuma City, located 80 km west of the epicenter. The disaster caused 180 deaths, damaged 5,542 homes and flooded 48% of Iwanuma’s land (Figure S2).

We conducted follow-up surveys of survivors four times in 2013, 2016, 2019 and 2022, gathering information on health status, lifestyle and housing types after the disaster. We inquired about PTG only in the 2022 survey.

The detailed flow chart of the analytic sample is presented in Fig. 1. The respondents with physical and cognitive disabilities at the baseline (n = 28) were excluded from this study because they are particulary vulnerable in the aftermath of disasters (Aldrich and Benson, Reference Aldrich and Benson2008), which may affect both PTS and PTG as confounders. Finally, the analytic sample was 1,489 respondents.

Figure 1.

Participant flow in this study.

The survey protocol was reviewed and approved by the Ethics Committees on Research of Human Subjects at the Harvard T.H. Chan School of Public Health, Tohoku University, Nihon Fukushi University and Chiba University. Informed consent was obtained at the time of survey collection.

Explanatory variables

PTS was assessed using the Screening Questionnaire for Disaster-Related Mental Health (SQDRMH) (Fujii et al., Reference Fujii, Kato and Maeda2007), originally developed and psychometrically validated in the aftermath of the 1995 Hanshin–Awaji earthquake in Japan. The instrument was specifically developed for screening disaster-related mental disorders of older survivors and has been psychometrically validated against the Japanese-language version of the Clinician-Administered PTSD Scale (CAPS) (Blake et al., Reference Blake, Weathers, Nagy, Kaloupek, Gusman, Charney and Keane1995).

In the surveys, we asked respondents: ‘People who have experienced the 2011 Japan Earthquake and Tsunami often report that their lives have changed dramatically, and they are constantly under various kinds of stress. Have you experienced any of the symptoms listed below in the past month?’, and then asked their symptoms using the following nine questions: ‘Do you have trouble falling asleep or sleeping through the night?’, ‘Do you have nightmares about the event?’, ‘Do you feel irritable?’, ‘Do you feel that you are hypersensitive to small noises or tremors?’, ‘Do you avoid places, people and topics related to the event?’, ‘Do you think about the event when you do not want to?’, ‘Do you have trouble enjoying things you used to enjoy?’, ‘Do you get upset when something reminds you of the event?’ and ‘Do you notice that you are making an effort to try not to think about the event or are trying to forget it?’.

The nine items exhibited a reasonable internal consistency (Cronbach’s α = 0.74). Following previous studies suggesting a non-linear association between PTS and PTG, we categorized PTS as binary: 1 for severely affected (6 to 9 points) and 0 for moderately or slightly affected (0 to 5 points) (Fujii et al., Reference Fujii, Kato and Maeda2007). Additionally, we used ordinal variables and, in a sensitivity analysis, a continuous variable for PTS.

Outcome variable

Our primary outcomes were obtained from objective assessments of functional, cognitive and physical disabilities, in addition to PTG.

1. 1. Functional disability. The level of functional disability was obtained from the Japanese Long-Term Care Insurance (LTCI) database. Since 2001, the Japanese government has implemented a national insurance scheme for older individuals requiring long-term care (e.g., home helpers). Applicants are classified into one of seven care levels through a standardized multistep assessment of functional and cognitive disabilities conducted by trained investigators and physicians. These levels correspond to the severity of their disabilities and the estimated hours of home care needed each week (e.g., bathing, dressing, cleaning the house and preparing meals) (Tamiya et al., Reference Tamiya, Noguchi, Nishi, Reich, Ikegami, Hashimoto, Shibuya, Kawachi and Campbell2011).

2. 2. Cognitive disability. Trained investigators assess applicants’ cognitive function (e.g., short-term memory, orientation and communication) and mental and behavioural disorders (e.g., delusions of persecution and confabulation) using a standardized protocol implemented during in-home assessment. Following this assessment, applicants are classified into one of seven levels, ranging from 1 (experiencing some cognitive deficits but otherwise almost completely independent) to 7 (requiring constant treatment in a specialized medical facility), based on the severity of their cognitive disability. Additionally, a panel of physicians independently assess the level of cognitive disability of applicants to determine their care requirements (Olivares-Tirado and Tamiya, Reference Olivares-Tirado, Tamiya, Olivares-Tirado and Tamiya2014).

3. 3. Physical disability. Under the LTCI scheme, trained investigators assess applicants’ activities of daily living and instrumental activities of daily living, and classify applicants into one of eight levels (1: Suffers from some level of disability but is able to function independently for the most part and can manage outings alone using public transportation, to level 8: Spends the whole day in bed and requires assistance in toileting, preparing meals, changing clothes and even turning over in bed).

4. 4. PTG. PTG was measured using the Short Form of the Expanded Version of the Posttraumatic Growth Inventory (PTGI-X-SF-J) (Oshiro et al., Reference Oshiro, Soejima, Kita, Benson, Kibi, Hiraki, Kamibeppu and Taku2023). This scale assessed five dimensions of PTG: improved relationship with others, increased personal strength, identification of new possibilities, positive spiritual changes and increased appreciation of life.

We instructed participants ‘Please tell us to what extent your daily life has changed as a result of experiencing the 2011 Japan Earthquake and Tsunami’ and asked them the following questions. Improved relationship with others was measured using the following two items: ‘I have a greater sense of closeness with others’ and ‘I more clearly see that I can count on people in times of trouble.’ Increased personal strength was measured using the following two questions: ‘I know better how to handle difficulties’ and ‘I discovered that I’m stronger than I thought I was.’ The following two items were used to assess the identification of new possibilities: ‘New opportunities are available which wouldn’t have been otherwise’ and ‘I established a new path for my life.’ Positive spiritual changes were measured using the following two questions: ‘I feel more connected with all of existence’ and ‘I have a greater sense of harmony with the world.’ We used the following two questions to measure the increased appreciation of life: ‘I have a greater appreciation for the value of my own life’ and ‘I can better appreciate each day.’ Responses were ordered along a six-point Likert scale (1: not at all, 6: very strong).

As shown in Figure S3, we implemented a confirmatory factor analysis to check the construct validity of PTG measured by 10 items and found a reasonable five-factor solution with acceptable goodness-of-fit indices (comparative fit index 0.981, root-mean-square error of approximation 0.079 and standardized-root-mean-square residual 0.025). We calculated the arithmetical mean of the responses to the 10 items to create a total PTG score, which indicated good internal consistency (Cronbach’s α = 0.94).

Covariates

Based on previous studies (Hikichi et al., Reference Hikichi, Aida, Kondo and Kawachi2021, Reference Hikichi, Aida, Kondo, Tsuboya, Matsuyama, Subramanian and Kawachi2016), we chose several baseline demographic variables as potential confounding variables for the association of PTS and PTG: age, sex, equivalized household income, educational attainment, divorce or bereavement and employment status.

For equivalized household income, we divided the household’s gross income by the square root of the number of household members (OECD, 2013).

Based on previous findings about risk factors for PTG, we also adjusted for the number of adverse life events (Fraus et al., Reference Fraus, Dominick, Walenski and Taku2023) at the baseline. Participants were asked whether they had experienced any of the following events within the past year: resignation, living alone, financial difficulty, loss of a spouse, loss of relatives or friends and serious illnesses. We then totalled the number of adverse life events reported.

We also chose depressive symptoms, and smoking and drinking habits in each survey as comorbidities of PTS. Depressive symptoms were measured by the Geriatric Depression Scale-15 (GDS-15; Sugishita et al., Reference Sugishita, Sugishita, Hemmi, Asada and Tanigawa2017). The baseline GDS-15 items showed good internal consistency, with a Cronbach’s α of 0.79. We created a binary variable, dichotomized at a clinically validated cutoff point, to demonstrate that a higher level of depressive symptoms is correlated with PTG. This variable distinguished between lower risk (4 points and under) and higher risk (5 points and over) (Weintraub et al., Reference Weintraub, Oehlberg, Katz and Stern2006). In a sensitivity analysis, we utilized a continuous variable of the GDS-15 score.

We also considered post-disaster housing. Immediately after the disaster, survivors who lost their homes had options, including moving into disaster-relief housing, renting apartments or purchasing new housing. When the disaster-relief housing closed in April 2016 and Iwanuma City opened permanent housing, survivors had another opportunity to choose from government housing, market rentals or home purchases. The variable indicates four housing categories: 1 = no relocation, 2 = government-provided housing, 3 = apartments in the open rental market and 4 = newly purchased housing.

Statistical analysis

First, we employed linear regression models to examine the cross-sectional association of PTG and PTS with functional, cognitive and physical disabilities considering baseline (pre-disaster) covariates.

Second, we examined the association between PTS and PTG using three models: (1) adjusting for only baseline covariates, (2) including time-series variables and (3) considering time-dependent confounding and inverse probability weighting (i.e., marginal structural models). In this analysis, we utilized two types of datasets: a five-wave panel (Fig. 1) and imbalanced data consisting of baseline respondents who participated in at least one of the follow-up surveys. Specifically, the imbalanced data included a significant number of attrition cases, which were adjusted for in the analysis.

We estimated the stabilized inverse probability of receiving the treatment (i.e., a higher risk of PTS) SW(t) and the probability of participating in surveys up to time t SW(c) to create a pseudo-population to balance the distribution of potential confounders across exposure levels and remained cases (Godin et al., Reference Godin, Elbejjani and Kaufman2012). SW(t) and SW(c) are defined as

\begin{equation*}SW\left( t \right) = \mathop \prod \limits_{k = 0}^t \frac{{f\left\{ {{\textrm A}\left( k \right){\text{|}}\bar A\left( {k - 1} \right),\,V,\,\bar C\left( k \right) = 0} \right\}}}{{f\left\{ {{\textrm A}\left( k \right){\text{|}}\bar A\left( {k - 1} \right),\bar L\left( k \right),\,\bar C\left( k \right) = 0\,} \right\}}}\end{equation*}

and

\begin{equation*}SW\left( c \right) = \mathop \prod \limits_{k = 0}^t \frac{{{\text{Pr}}\left\{ {C\left( k \right) = 0{\text{|}}\bar C\left( {k - 1} \right) = 0,\,\bar A\left( {k - 1} \right),\,V} \right\}}}{{{\text{Pr}}\left\{ {C\left( k \right) = 0{\text{|}}\bar C\left( {k - 1} \right) = 0,\bar A\left( {k - 1} \right),\bar L\left( {k - 1} \right)}\! \right\}}}\end{equation*}

where ${\textrm A}\left( k \right)$ denotes the exposure at year $k$, $\bar A\left( {k - 1} \right)$ represents the exposure history prior year $k$, $V$ is baseline covariates, $\bar L\left( {k - 1} \right)$ represents the covariates history including $V$, $C\left( k \right)$ is the incident attrition at year $k$ and $\bar C\left( {k - 1} \right)$ is participating history until year $k - 1$.

We estimated only $SW\left( t \right)$ for the five-panel data because of no attrition cases, while we calculated both $SW\left( t \right)$ and $SW\left( c \right)$ for the imbalanced data to adjust selection bias due to dropped cases.

Throughout these analyses, we utilized binary and ordinal variables for PTS as previous studies have suggested non-linear associations between PTS and PTG (Kadri et al., Reference Kadri, Gracey and Leddy2022). Specifically, the ordinal variable of PTS was likely a suitable choice to understand their unique association, especially since respondents with severe PTS exhibited higher proportions of comorbidities, such as severe depressive symptoms and current smoking (Table S1). In a sensitivity analysis, we used a continuous variable for PTS.

We implemented multiple imputation for missing values using the Markov chain Monte Carlo method that assumes missingness at random. Twenty datasets were created, and each result of analyses was combined using the Stata command ‘mi estimate’. All analyses were performed using STATA version 17.0 (STATA Corp LP, College Station, TX, USA).