Survival Analysis for Cohort of Bhopal Gas Disaster Victims during 1985-2015

Abstract

Context:

After the gas tragedy on the night of December 2/3, 1984, at Bhopal, the Indian Council of Medical Research (ICMR) started following up on four population cohorts with different levels of post-disaster mortality from December 3–6, 1984.

Aims:

The present study was undertaken to estimate the survival time of the cohort, and investigate the risk of mortality based on exposure, gender, and median age.

Settings and Design:

Survival analysis is generally used to evaluate factors associated with the time to an event of failure or death among any covered population.

Methods and Materials:

To know the cause of death and mortality rate, a retrospective cohort analysis was conducted on the outcomes of 92,320 individuals with an exposed and non-exposed group from 1985 to 2015 in Bhopal, India.

Statistical Analysis Used:

Basic survival analysis method, Kaplan–Meier method, and Cox proportional hazard regression model were used to analyze the mortality risk.

Results:

During the past 30 years, the survivability was 87.25%, and the mortality rate was 7.2% for the cohort population of Bhopal gas survivors. Cox regression analysis showed that exposed, males, and individuals above 21 years (at the time of the disaster) were at higher risk of mortality from 1985 to 2015.

Conclusions:

During the initial two phases, the mortality was higher in the exposed group, but over time, their survival turned out to be the same in both groups.

INTRODUCTION

Toxic gases mainly containing methyl-isocyanate (MIC) leaked on the intervening night of December 2/3, 1984, from the Union Carbide Factory in Bhopal, Madhya Pradesh, India, and led to a chemical disaster, which is also known as the Bhopal gas disaster (BGD).[1,2,3,4] Exposure to toxic gases resulted in massive mortality and morbidity in the exposed population of Bhopal. About 2,500 people died immediately, and several thousand suffered from various ailments of the lungs, eyes, gastrointestinal tract, and skin due to varying quantum of exposure and for a variable duration.

Indian Council of Medical Research (ICMR) started a long-term population-based epidemiological study on the health effects of toxic gas exposures soon after the disaster, that is, in August 1985. In this study, ICMR started following up on population cohorts in different geographic areas of Bhopal city to keep a scientific account of the post-disaster burden of morbidity and mortality among the MIC-exposed population. The exposed geographic areas were classified as severely affected, moderately affected, and mildly affected according to the mortality rates accountable to toxic gas exposure experienced during 3–6 December 1984. The study was executed through six community health clinics. A medical officer and four research assistants (RA) headed each community health clinic. RA was to ask for and record all the information regarding morbidity, mortality, actual events, and pregnancy outcomes within the cohort since the last visit (6 months ago). A matching cohort from the unaffected or non-exposed population was also selected and followed up during these years for comparison purposes. The concerned have published the technical report parts I and II, respectively, analyzing the mortality rates and causes of death of the followed-up cohorts from 1985 to 1995 and from 1996 to 2010. The same cohort under long-term epidemiological study has been continuously followed up by ICMR- National Institute for Research in Environmental Health (NIREH), Bhopal since 2011.

The time, beginning from a well-defined point to the incidence of a specified event is called the survival time, and the analysis of such group data is the survival analysis. In community trials, the effect of an intervention over a long duration is considered by estimating the number of subjects who survived. Sometimes, it is remarkable to compare the survival of subjects in two or more interferences. In circumstances where existence is the issue, the variable of interest would be the length of time that passes before some event happens. By the way, after this disaster, a lot of efforts were made by the Madhya Pradesh State government as well as the Central government for providing better health facilities, economic enrichment, and environmental improvement for the BGD-exposed population.[5] In this context, survival analysis is an essential tool to estimate the impact of BGD and subsequent mitigation activities. In this study, we estimated and compared the mean survival time for the BGD-exposed and non-exposed population cohorts.

Published reports suggested that survivors of the BGD (exposed to MIC) are suffering from residual or chronic morbidities such as chronic obstructive pulmonary diseases, ocular morbidities, and chronic kidney diseases.[1,2,3] Respiratory ailments remained the most important cause of mortality among survivors during these years.[6,7,8,9,10] The ICMR population-based epidemiological study offers a unique opportunity to compare the mortality burden and causes of death between exposed and non-exposed populations. Although several reports such as annual reports, consolidated reports, and technical reports on cohorts during 1985–2015 by Bhopal Gas Disaster Research Centre (BGDRC), Center for Rehabilitation Studies (CRS) and ICMR-NIREH have been published, the voluminous scientific data of cohort study needs further in-depth analyses to estimate the survival time of the cohort. Thus, the present study is undertaken to estimate the survival time, and explore leading causes of death among exposed and non-exposed groups.

SUBJECTS AND METHODS

The epidemiological studies data from 1985 to 2015 were considered in the present study.[1,2,6,11] A master database was developed as a cohort for this purpose; including a total of 92,320 individuals who were present at home on the night of the tragedy. These individuals were assigned unique identification numbers (locality no., ICMR family no., ID no.), information on demographic variables, and morbidity/mortality variables were collected at 6 monthly intervals from 1984 to 2015 using a detailed questionnaire available in technical reports (I & II).[1,6]

For this survival analysis, the main outcome of interest is the “death” of cohort subjects and “Censored cases” due to the shifting of the population to different places caused by the marriage migration of females of the cohort. The analysis had mainly two objectives, that is, (i) to estimate the survival time of the cohort, and (ii) to investigate the risk of mortality based on the exposure (exposed with non-exposed), gender (male with female), and median age (up to 20 years with 21 years and above) of the cohort.

The survival function is defined as the probability that an individual survives at time t, and censoring occurs in time-to-event data (the time from a defined origin until the event of interest) when the event has not been observed (i.e., the time to the event is unknown). For the first objective, a non-parametric method, the Kaplan–Meier method,[12] also known as the “product-limit method,” is used to estimate the probability of survival past a given time point (i.e., it calculates a survival distribution). Furthermore, the survival distributions of two or more groups of a between-subjects factor can be compared for equality. The outcome measure is “time to event”; so, we used the Kaplan–Meier analysis for univariate analysis and the Cox proportional hazard (PH) model for multivariate analysis. After fitting the Cox PH model, the three categorical variables, namely, exposure (exposed and non-exposed), gender (male and female), and age group (up to 20 years and above to 21 years based upon median age group at the time of the disaster) were included in the model. We have tested whether any of the variables grossly violates the assumption of proportionality of hazards (which must be met). To describe how the factors jointly impact survival, a multivariate Cox regression analysis was performed; the output above shows the regression beta coefficients, the effect sizes (given as hazard ratios), and statistical significance for each variable concerning overall survival. Mathematically, the Cox PH model is written as the risk function:

where, h₀(t) is the benchmark risk function and X (x1, x2,…, xm), and β (b1, b2,…, bm) are the observed variables and their regression coefficients, respectively. Here “t” in h(t) reminds us that the hazard may (and probably will) vary over time. The Cox PH model is essentially a multiple linear regression of the logarithm of the hazard on the variables xi. The baseline hazard is an “intercept” term that varies with time. The covariates then act multiplicatively on the hazard at any point in time. This provides us with the key assumption of the PH model: the hazard of the event in any group is a constant multiple of the hazard in any other. This assumption implies that the hazard curves for the groups should be proportional and cannot cross. The inspection level α was set to 0.05 (two-sided). A positive regression coefficient indicates an increased hazard of death, and a negative regression coefficient indicates a lower hazard. The regression coefficients were presented with their standard error (SE), which measures the regression coefficient's uncertainty. The IBM-SPSS Statistics-25 software was used for statistical analyses of the data. Also in this study, the primary cause of death was coded according to ICD-10, and percentage distribution was presented during the initial and later phase of the disaster. The information on causes of death was included since 1986: the cause of death analysis is based on 6,160 death records categorized separately during 1986–2000 and 2001–2015.

Ethical clearance

Data analysed in the current manuscript pertains to “Population based long term epidemiological studies on health effects of Bhopal toxic gas exposure” (Project No. 02) which was approved by Ethical Committee of Bhopal Gas Disaster Research Centre, Indian Council of Medical Research, Gandhi Medical College, Bhopal on 29^th May 1991.

RESULTS

During the study period, a total of 6,609 individuals died, and overall the mortality rate was 7.2%. The median age of these individuals was 40 (25, 55) years, ranging from 1 to 99 years; 5,540 were in exposed areas, and 1,159 were in non-exposed areas. In Table 1, the percentage distribution of age and sex profile of the individuals in the exposed and non-exposed areas of the cohort is given. From Table 2, during the 365 months study period, the estimated mean survival time was about 344.5 months for the exposed group and about 349.5 months for the non-exposed group, which shows that there was about 5 months survival time difference between the exposed and non-exposed groups. From the log-rank test, the survival of the exposed and non-exposed groups was significantly different (P- < 0.05). There was a significant difference in the estimated survival time for males and females, which were about 343.5 and 347.4 months, respectively. Similarly, for the cohort below 20 years and above 21 years, the estimated survival time was approximately 358 and 330 months, respectively. From annexure-1, the probability of survival in 365 months was 0.8741, that is, 87.41% of individuals survived from August 1985 to December 2015.

Table 1:

Percentage distribution of age–sex by exposure of the cohort population

Gender	Median age	Exposed		Non-exposed
		n	%	n	%
Male	Up to 20 years	21193	52.8	4330	51.3
	21 years and above	18942	47.2	4107	48.7
Female	Up to 20 years	19996	54.8	3992	55.2
	21 years and above	16523	45.2	3237	44.8

Table 2:

Estimate of the survival time (in months) using the Kaplan–Meier method

Population/exposure	Total n	No. of Events	Censored		Mean survival time			Std. Error	Chi-square	d. f.	P
			n	%	Estimate	LB	UB
Overall cohort	92320	6609	85711	92.8	345.4	344.9	345.9	0.25	-	-	-
Exposure
Exposed	76654	5540	71114	92.8	344.5	343.9	345.0	0.29	44.22	1	<0.001
Non-exposed	15666	1069	14597	93.2	349.5	348.4	350.5	0.52
Gender
Male	48572	3829	44743	92.1	343.5	342.8	344.2	0.36	94.89	1	<0.001
Female	43748	2780	40968	93.6	347.4	346.7	348.1	0.34
Median age
Up to 20 years	49511	1252	48259	97.5	357.9	357.5	358.4	0.21	3867.1	1	<0.001
21 Years and above	42809	5357	37452	87.5	330.2	329.3	331.1	0.47

From Figure 1(a), it is clear that at the end of 365 months, the probability of survival for the exposure group was approximately 0.875, that is, 87.5%, whereas for the non-exposed group, it was 0.895, that is, 89.5%. It can be noted that the mortality rate was very high during the initial phase, which later on became stable over time. Both survival lines originated from the same point; the slopes between the start and end points were 2.399 versus 1.561 from August 1985 to May 1994, 1.611 versus 1.448 from January 1996 to June 2010, and 0.979 versus 0.973 from January 2012 to December 2015 in non-exposed versus exposed areas. Figure 1(b) reflects that overall survival differences between males and females were approximately 2%, and Figure 1(c) shows a drastic decrease in the survival of people above 20 years of age. Among that cohort, only 77.5% of individuals survived at the end of 365 months.

Figure 1:

Kaplan-Meier survival curves of the BGD cohort. (a) KM- survival curves by Exposure: Exposed-Non-exposed (ref) (b) KM- survival curves by Gender: Male- Female (ref) and (c) KM- survival curves by age group: Age 21 years & above- Up to 20 years (ref)

Kaplan–Meier survival curves of each group of categorical variables indicated that the model hypothesis might not be satisfied. The beta regression coefficients obtained in the studied models are all positive (b > 0) with statistical significance, suggesting that all factors included in the model influence the event speed (death). The hazard ratio (HR), that is, the risk of death is obtained from the exponential regression coefficient and gives the effect size of the predictors. Here, a multivariate Cox model was performed to describe the risk factors associated with a 365 months survival. From Table 3 it is clear that exposed (ref non-exposed), male (ref female) and median age up to 20 years (ref 21 years and above) variables had highly significant positive coefficients. The P value for all three overall tests (likelihood, Wald, and log-rank) was found to be significant. These tests evaluate the omnibus null hypothesis that all the betas (β) are positive. The test statistics are in close agreement in the above example, and the null hypothesis is rejected. The P value for the exposed group was <0.001, indicating a strong relationship between exposure level with the risk of death with exposure to gas.

Table 3:

Multivariate analysis for hazard ratios from the Cox PH model for the BGD cohort

Variables	B	SE	Wald	d.f.	Sig.	Exp (B)
Exposed (ref-non-exposed)	0.265	0.033	62.906	1	<0.001	1.304
Male (ref-female)	0.228	0.025	83.858	1	<0.001	1.256
>20 years (ref-<20 years)	1.730	0.031	3032.520	1	<0.001	5.640

Harrell's Concordance Index=0.698 (se=0.003). Likelihood ratio=4151 on 3 d.f., P≤0.001, Wald test=3180 on 3 d.f., P≤0.001, and Score (log-rank) test=4014 on 3 d.f., P≤0.001

Similarly, the P value for gender and the median age was <0.001, with an HR of 1.26 and 5.64, respectively, indicating a strong relationship between the gender and age group and increased risk of death in the male and above median age group. Thus, those exposed to MIC are at a 1.30 times higher risk of death than those not exposed, adjusting for gender and age group. Similarly, males had 1.26 times, and individuals above the median age of 21 had a 5.64 times higher risk of death than females and individuals up to the median age of 20 years, respectively. From Table 4 between 1986 and 2015, the most common cause of death was respiratory disorders, followed by cerebrovascular and digestive disorders in exposed as well as unexposed cohorts. The proportional mortality due to respiratory diseases was higher in the exposed cohort.

Table 4:

Primary cause of death in exposed and non-exposed areas during 1986-2015

ICD-10 code	Primary cause of death	1986-2000				2001-2015
		Exposed (2582)		Non-exposed (423)		Exposed (2562)		Non-exposed (593)
		n	%	n	%	n	%	n	%
G00 to G99	Nervous system	112	4.3	7	1.7	132	5.2	21	3.5
I00 to I99	Circulatory system	86	3.3	13	3.1	193	7.5	50	8.4
J00 to J99	Respiratory system	1211	46.9	108	25.5	1330	51.9	145	24.5
K00 to K93	Digestive system	244	9.5	49	11.6	229	8.9	51	8.6
R00 to R99	Symptoms, signs and abnormal clinical, fever.	89	3.4	58	13.7	89	3.5	88	14.8
S00 to T98	Injury, poisoning and certain other consequences of external causes	118	4.6	44	10.4	113	4.4	67	11.3
---	Senility	106	4.1	42	9.9	126	4.9	94	15.9
---	Other cause	124	4.8	49	11.6	274	10.7	33	5.6
---	Cause unknown	492	19.1	53	12.5	76	3.0	44	7.4

DISCUSSION

The BGD is undoubtedly an example of the worst industrial disaster leading to uncountable adverse health consequences ranging from immediate death to toxico-genomic effects.[13] The magnitude of adverse health consequences demands a complete and scientific understanding of the residual effects on the survivor population. In this study, mortality data from survivors of toxic gas leakage in the 1984 Bhopal gas tragedy and a cohort of the non-exposed population, representing India's general population in an urban setting are presented.

After three decades, the BGD survivor population has entered the age of adulthood to old age, that is, above 30 years to the age of 99 years. The population is also now vulnerable to the health concerns of old age, such as cardiovascular diseases, stroke, diabetes, and cancers.[12,14,15,16] In this study, it was observed that throughout follow-up years, there was an increase in the proportion of deaths due to diseases of the circulatory system and nervous system in both groups.

In the phase-I (0 to 100 months) and phase-II (101 to 300 months), the survival of the exposed population (slope 1.561 and 1.448, respectively) was lower as compared to the non-exposed population (slope 2.399 and 1.611, respectively), whereas in Phase-III (301 to 400 months), the survival pattern was almost equal in both the groups (slope in non-exposed was 0.979 and in exposed was 0.973). It is observed that during the initial two phases, the mortality rate was high in exposed areas during the initial phase and was stable in a recent period with reference to non-exposed areas [Figure 1a].

As reported in Table 2, significant differences were observed in the mean survival time for different population subgroups viz. exposure to BGD, gender, and age at the time of BGD (5, 4, and 28 months, respectively). The elderly population is more vulnerable to cardiorespiratory challenges. They have a lesser capacity to regenerate as compared to younger populations. This physiological difference between young and older (i.e., age 21 years and above) populations can explain the wide observed difference of 28 months in their survival time.[18] Longer survival time (of 4 months) among women may be due to biological factors such as genetics or hormones, which is well documented in the literature.[19] The shorter survival time among the exposed population could be attributed to the higher burden of respiratory morbidity that is also reported in the literature.[1,17] Health facilities, economic enrichment, and socio-environmental improvement provided by the State, as well as the Central government, to the BGD survivors might have positively influenced their survival.[5]

According to the report, the burden of premature mortality mainly includes deaths among those aged 30 to 69 years i.e. before the age of 70 years, which claimed 65% of India's 10.3 million deaths in 2014.[20,21] Office of the Registrar General and Census Commissioner, under the Ministry of Home Affairs, the Government of India, has published a detailed report based on the Verbal Autopsy of 1,82,827 deaths during 2010–2013 across India.[22,23] According to the report, the burden of premature mortality mainly includes deaths among those aged 30 to 69 years.[19] With the current mortality rates, India is predicted to account for the highest number of (9.26 million) premature deaths by 2030. As per previous reports, respiratory diseases were the most critical contributor to morbidity and mortality among the exposed population.[1,12,17]

METHODOLOGICAL CONSIDERATION

One of the strengths of this study is that it presented the data of the exposed population with the data matching the non-exposed cohort, allowing us to compare the findings to conclude. One of the most challenging tasks in operating a cohort study is to hold the cohort, and this study too suffered a loss of the original cohort. This is one of its kinds of cohorts followed up for this long time, and it also has its limitations. Also, there were many time methodical changes in socioeconomic observations and data collection frequency. The causes behind losses include shifting the population to different places, marriage migration, shifting of the young age cohort to older age one, and subsequent removal by death. The present study involved field-based verbal autopsies of the deaths reported during a routine epidemiological survey where self-reported causes of death were verified with available documentary reports. Such methodologies sometimes encounter limitations related to recall or reporting.

CONCLUSIONS

Survival analysis showed that MIC-exposed males from the older age group were at high-risk for mortality during the period. The mortality rate was high in exposed areas during the initial phase and stable in the recent period with reference to non-exposed areas. The proportion of death due to respiratory diseases was higher in exposed areas as compared to non-exposed areas throughout the study duration. In recent years, as the cohort is aging, the proportion of deaths due to circulatory and nervous system diseases is increasing in both groups.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.

Annexure 1: Survival life table of the cohort population from 1985 to 2015.

Visit	Month-Year	Number Entering Interval	Number Withdrawing during Interval	Number Exposed to Risk	Number of Terminal Events	Cumulative Proportion Surviving at End of Interval
0	Aug-85	92320	1413	91613.5	0	1.0000
1	Aug-85 to Dec-86	90907	1103	90355.5	191	0.9979
2	Jan-86 to Jun-86	89613	1870	88678.0	171	0.9960
3	Jul-86 to Nov-86	87572	9286	82929.0	266	0.9928
4	Dec-86 to Nov-87	78020	675	77682.5	130	0.9911
5	Dec-87 to May-88	77215	1333	76548.5	61	0.9903
6	Jun-88 to Nov-88	75821	741	75450.5	56	0.9896
7	Dec-88 to May-89	75024	1674	74187.0	259	0.9861
8	Jun-89 to Nov-89	73091	1599	72291.5	133	0.9843
9	Dec-89 to May-90	71359	1240	70739.0	127	0.9825
10	May-90 to Nov-90	69992	1881	69051.5	140	0.9806
11	Dec-90 to May-91	67971	1359	67291.5	110	0.9789
12	Jun-91 to Nov-91	66502	584	66210.0	95	0.9775
13	Dec-91 to May-92	65823	1046	65300.0	87	0.9762
14	May-92 to Nov-92	64690	1268	64056.0	71	0.9752
15	Dec-92 to May-93	63351	543	63079.5	112	0.9734
16	May-93 to Nov-93	62696	738	62327.0	233	0.9698
17	Dec-93 to May-94	61725	1986	60732.0	63	0.9688
18	Jan-96 to Jun-96	59676	970	59191.0	135	0.9666
19	Jul-96 to Dec-96	58571	966	58088.0	90	0.9651
20	Jan-97 to Jun-97	57515	568	57231.0	123	0.9630
21	Jul-97 to Dec-97	56824	811	56418.5	81	0.9616
22	Jan-98 to Jun-98	55932	573	55645.5	128	0.9594
23	Jul-98 to Dec-98	55231	502	54980.0	47	0.9586
24	Jan-99 to Jun-99	54682	504	54430.0	109	0.9567
25	Jul-99 to Dec-99	54069	610	53764.0	109	0.9547
26	Jan-00 to Jun-00	53350	697	53001.5	172	0.9516
27	Jul-00 to Dec-00	52481	------	52481.0	104	0.9497
28	Jan-01 to Jun-01	52377	353	52200.5	165	0.9467
29	Jul-01 to Dec-01	51859	914	51402.0	98	0.9449
30	Jan-02 to Jun-02	50847	445	50624.5	164	0.9419
31	Jul-02 to Dec-02	50238	643	49916.5	88	0.9402
32	Jan-03 to Jun-03	49507	742	49136.0	132	0.9377
33	Jul-03 to Dec-03	48633	684	48291.0	73	0.9363
34	Jan-04 to Jun-04	47876	596	47578.0	105	0.9342
35	Jul-04 to Dec-04	47175	532	46909.0	61	0.9330
36	Jan-05 to Jun-05	46582	904	46130.0	130	0.9304
37	Jul-05 to Dec-05	45548	473	45311.5	65	0.9290
38	Jan-06 to Jun-06	45010	469	44775.5	118	0.9266
39	Jul-06 to Dec-06	44423	411	44217.5	63	0.9253
40	Jan-07 to Jun-07	43949	558	43670.0	106	0.9230
41	Jul-07 to Dec-07	43285	188	43191.0	72	0.9215
42	Jan-08 to Jun-08	43025	463	42793.5	120	0.9189
43	Jul-08 to Dec-08	42442	561	42161.5	86	0.9170
44	Jan-09 to Jun-09	41795	602	41494.0	153	0.9136
45	Jul-09 to Dec-09	41040	500	40790.0	113	0.9111
46	Jan-10 to Jun-10	40427	1617	39618.5	112	0.9085
47	Jul-10 to Jul-11	38698	2193	37601.5	143	0.9051
48	Jan-12 to Dec-12	36362	697	36013.5	244	0.8989
49	Jan-13 to Jun-13	35421	525	35158.5	142	0.8953
50	Jul-13 to Dec-13	34754	484	34512.0	165	0.8910
51	Jan-14 to Jun-14	34105	656	33777.0	104	0.8883
52	Jul-14 to Dec-14	33345	564	33063.0	144	0.8844
53	Jan-15 to Jun-15	32637	1729	31772.5	116	0.8812
54	Jul-15 to Dec-15	30792	30668	15458.0	124	0.8741