2.1 Study population

LASI is a longitudinal survey (Wave 1, 2017−18) that provides data on many morbidities in individuals 45 years and above and their spouses irrespective of age and their socio-economic and demographic details, covering 28 states and 8 UTs of India. LASI has adopted a multi-stage stratified areas probability cluster sampling design, with three stages in the rural and four stages in the urban areas, respectively. Firstly, a primary sampling unit (PSU) was selected from each state/union territory (UT), followed by a village (from rural) or ward (from urban) area in the second stage. Finally, households were selected from the rural areas. However, in urban areas, census enumeration blocks (CEB) were selected randomly from each urban area, after which households were chosen from the selected CEB (International Institute for Population Sciences (IIPS), 2020). The present study utilised merged information from individual and biomarker datasets. The dataset contained samples of 73,396 adults aged 45 years and above, and in this research work, we primarily analysed data from 66,606 individuals after excluding the missing data.

2.2 Data availability and ethical consideration

Prior informed consent (written and verbal) from all the participants was collected by the field survey agencies. LASI administered consent forms at household and individual levels, following the Human Subject Protection [26]. The Indian Council of Medical Research (ICMR) extended the necessary guidance and mandatory ethical approval for conducting the LASI survey. The Institutional Review Board at IIPS provided additional approval for the study protocol. All methods were carried out under relevant guidelines and regulations by the ICMR.

2.3 Variables definitions

2.4 Statistical method

First, we computed the weighted prevalence distribution of morbidities and multimorbidity of the target population. Second, we plotted the relative proportional share of morbidities over age and calculated the weighted age-specific relative prevalence share of multimorbidity, single morbidity and zero morbidity. Third, we hypothesized that the distribution of Multimorbidity may be similar or different by the covariates; for this purpose, the Chi-square test was employed to examine differences in the distribution of multimorbidity across socioeconomic, demographic and lfestyle factors. Finally, the susceptibility to multimorbidity was assessed using the CART model, based on a set of dynamic exogenous determinants. Further, all p-values were two-tailed, and differences were considered statistically significant at p < 0.05.

2.4.1 CART model.

The Classification and Regression Tree model (CART) is a method of regression or classification trees and used as a growing method for the decision tree, run for the one outcome variable (Multimorbidity, Single morbidity and Zero morbidity).

The CART model splits the data with maximum homogeneity within the node. The degree of a non-homogeneous subset in a node, is an indication of impurity. In selecting the best splitter, CART attempted to maximize the average purity of the two child nodes. The way to measures of purity can be chosen freely, can be called a splitting criterion, or splitting functions. For this study, this impurity is measured using the Gini index. Calculation of the Gini index was obtained by the below formula

Where P (i | t) was the relative frequency of class i at node t, and c was the number of classes.

These calculations will achieve the highest value if the distribution of uniform class and has the smallest value if it contains all the records with the same class. Splitting of the data continues until homogeneity or stopping criteria are met in the node [28]. The minimum decrease in impurity was set at 0.0001. The growth limit was predetermined on a maximum tree depth of 5 and minimum leaf node observation of 5. Missing values were treated as missing values.

Three distinct strengths of CART that make it particularly applicable for analysing complex diseases models [29]. First, the hierarchical structure of CART models is often more intuitive than traditional regression models because it mimics the heuristics of decision making with conditional risks. Second, CART can outperform standard regression models when predicting outcomes in the presence of nonlinear relationships and interactions thresholds may vary with direct and intermediating risk factors. For example, the risk of diagnosis with CVDs in the individuals decreases if they have later onset of cholesterol and increases in the case of early onset of cholesterol. Third, CART affords the data greater freedom to speak for themselves. Whereas regression models are refined by comparing across a limited number of possible specifications, CART performs an exhaustive search over all possible cut-points and predictors [30]. As a result, the precise form of the relationship between a predictor and outcome is not delimited by the inclusion/exclusion of higher order terms.

All analyses were performed with SAS version 9.4.M8, R Studio version 2024.04.2−764 and STATA 17. A. The CART model was constructed using the “rpart” package in R and the resulting decision tree was visualized using the “rpart.plot” package.

3.1 Prevalence, distribution and patterns of multimorbidity and morbidities

Table 1 shows the burden of morbidities among the older adult population aged 45 and above in India. Zero morbidity prevalence was 25.29%, single morbidity prevalence was 23.78%, and multimorbidity was 50.90%. Amongst 36 diseases, hypertension showed the highest burden of 26.72%, closely followed by myopia (24.2%), and bone-joint & other diseases (0.01%) had the lowest prevalence. Most chronic morbidities showed prevalence in the range of 1–7%, with a mean value of 5%. Therefore, morbidity with a high prevalence of hypertension, myopia, hypermetropia (20.75%), gastrointestinal condition (17.98%), cataracts (14.18%), and diabetes (12.21%) confirmed a larger morbidity burden in the Indian population compared to other morbidities. It indicates the predominance of few acute and degenerative diseases for, unhealthy aging among older adults in India.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Prevalence distribution of causes and thirty-six morbidities and conditions among older adults, LASI Wave-1 (2017−18), India.

https://doi.org/10.1371/journal.pone.0323890.t001

Fig 1 illustrates the relative age-specific distribution of morbidity composition among adults aged 45 years and above. The stacked area chart reveals that a limited number of non-communicable diseases, particularly hypertension, diabetes, and cataract, constitute a substantial proportion of the total morbidity burden across all age groups. While the relative contribution of hypertension and diabetes slightly declines in older age cohorts, conditions such as cataract and hearing disorders become more prominent, reflecting age-associated degenerative changes. Less prevalent morbidities, including congenital, psychiatric, and skin-related conditions, consistently contribute minimally across the lifespan. However, the relative share of severe diareses like tuberculosis, cancer and heart diseases in all groups remained significantly small and constant across all age groups.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Age-specific relative share of thirty-six morbidities and conditions, LASI Wave-1 (2017−18), India.

https://doi.org/10.1371/journal.pone.0323890.g001

Fig 2 depicts the relative prevalence share of Multimorbidity, Single morbidity and Zero morbidity over the age groups. The share of multimorbidity expanded by 20.37% when comparing individuals aged 45–49 (43.47%) to those aged 80–84 years (63.84%), with relative percentage change of 46.87. This indicates that as the population ages, particularly into the oldest age group, they are significantly more likely to experience multiple chronic conditions. On the contrary single morbidity proportion remained constant and the relative share remained plateaued across all age groups, while zero morbidity share shirked in later ages, this reduction was largely offset by a corresponding rise in the burden of multimorbidity in later age-groups among the older adults in India.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Age-specific relative proportional distribution of Zero morbidity, Single morbidity and Multimorbidity, LASI Wave-1 (2017−18), India.

https://doi.org/10.1371/journal.pone.0323890.g002

Further, in Fig 3,the stacked bar plot illustrates a pronounced age-related shift   within the multimorbidity burden among older adults in India. The prevalence of zero morbidity steadily declines from 32.98% at ages 45–49 to 15.17% among those aged 85+, reflecting a reduction of over 17 percentage points and indicating diminishing health with advancing age. Single morbidity shows a modest decline from 23.55% to 20.58%, suggesting a shift toward more complex health conditions in older age. In contrast, the proportion of individuals with 2–4 morbidities increase consistently from 38.86% in the 45–49 group to 50.74% in the 75–79 group, followed by a slight decline to 50.27% in the 85+ group, suggesting a high but plateauing burden in the oldest-old. Similarly, >4 morbidities rise markedly from 4.61% at ages 45–49 to a peak of 16.96% in the 70–74 group, before declining slightly to 13.98% at 85+. These patterns reveal that the overall burden of multimorbidity intensifies with age, predominantly driven by individuals with 2–4 morbidities; however, a notably steep increase is also observed in the prevalence of >4 morbidities, particularly up to the mid-70s. Interestingly, a modest decline or plateau in the oldest age groups suggests the potential influence of survivorship bias or selective mortality, wherein individuals with the highest disease burden may not survive into advanced old age.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Age-specific relative proportional distribution of multimorbidity counts, LASI Wave-1 (2017−18), India.

https://doi.org/10.1371/journal.pone.0323890.g003

Table 2 presents the weighted percent distribution of the three categories of morbidity status, by background characteristics in the Indian older adults population aged 45 years and above. The background variables were categorised as predisposing, enabling, and risk factors. Among the predisposing factors, women exhibited a slightly higher prevalence of multimorbidity (51.86%) compared to men (49.85%), indicating a marginal gender disparity in disease burden. By place of residence, multimorbidity was 62.99% in the urban population compared to 45.4% in the rural population. Social groups are of differential importance in Indian society; among social groups, the upper caste category showed a large share of 60.19% in the multimorbidity category. On the other hand, the most socially vulnerable STs category showed a small share of 31.12% in the multimorbidity category. It indicates a lack of medical infrastructure can be a factor for assessing medicine and drug facilities as when required in a society.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Descriptive characteristics of Zero morbidity, Single morbidity and Multimorbidity by exogenous facors, LASI Wave-1 (2017−18), India.

https://doi.org/10.1371/journal.pone.0323890.t002

Amongst religions, Hindus with the largest base showed a share of 49.71%, Christians with the smallest sample size showed a share of 47.71%, and others with a small sample size showed the largest share of 63.42% in the multimorbidity category. By marital status, the share of multimorbidity was 57.03% in widowed, followed by a share of 49.34% in the currently married person category. The share of multimorbidity was higher in persons living alone (56.18%) than in persons living with spouses (49.35%). Also, persons living with disability or impairment show a larger share in the multimorbidity category compared to normal persons. The predisposing factors explain a large share of the Indian population in the multimorbidity category, except in the young-adult age group of 45–49 years.

Amongst enabling factors, the share of multimorbidity was 60.63% in persons with schooling for more than ten years versus 44.49% in persons without schooling. As measured by the MPCE class, the affluent class showed a large share of 62.93% in the multimorbidity category, whereas the poorest class showed a share of 39.87%. By work status, persons currently not working showed a larger share of 60.56% in the multimorbidity category compared to those currently working with 43.82%.

Childhood health has long-term implications in the later years of an individual’s life. Individuals who had very poor childhood health endured a large share of 75.34% in the multimorbidity category in comparison to individuals who had better childhood health showing a share of ~50%. Amongst better childhood health conditions, persons with very good childhood health statuses showed a share of 51.74% in multimorbidity and the remaining 24.21% in single morbidity and 24.05% in zero morbidity. It implies that serious illness in childhood is the early origin of multimorbidity. A similar distribution of morbidity status was also apparent for self-rated health status. Individuals with poor self-rated health status reported a large share of 70.22% in the multimorbidity category compared to persons with good self-rated health status showing 38.37%. It points out an affirmative relationship between low self-rated health status and the prevalence of multimorbidity in the population. By region, the western, northern, and southern regions showed a larger percentage of the population enduring multimorbidity compared to the eastern, central, and north-eastern regions. Regional diversity in multimorbidity is expected as many states of India experience variations in weather and monsoons, which affects the level of mortality and morbidity. Most of these enabling factors reveal a large proportion of multimorbidity in populations with lower levels of health conditions and lower socio-economic status compared to their counterparts.

A large percentage of the obese (71.52%) and overweight (61.79%) population had multimorbidity; contrarily, a small percentage of the underweight population (39.61%), another facet of malnutrition, had multimorbidity. Finally, by applying the chi-square test of independence concludes a statistically significant association between the outcome variable (morbidity status) and covariates, as the P-value at the 5% level of significance was less than 0.001 for each covariate.

3.2 Conditional role of exogenous factors on multimorbidity susceptibility

To further identify the optimal leading covariates and their joint conditional role for multimorbidity susceptibility among older adults in India, we constructed the CART model. Fig 4 depicts the conditional risk and sample size of each of the three outcomes in each box respectively: Zero morbidity (right digit), Single morbidity (middle digit) and Multimorbidity (left digit). CART model deducted that childhood health, place of residence, age, BMI, caste category and education level are the leading optimal covariates (Table 3) for multimorbidity risk among older adults in India.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Leading factors of multimorbidity susceptibility, LASI Wave-1 (2017−18), India.

https://doi.org/10.1371/journal.pone.0323890.t003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. CART tree with hierarchical condition risk for multimorbidity susceptibility by exogenous factors, LASI Wave-1 (2017−18), India.

https://doi.org/10.1371/journal.pone.0323890.g004

The leftist terminal node (a) showed the highest risk of multimorbidity (0.57) was among those individuals who had moderate or poor childhood health. This risk probability is followed by the insignificant differential of risk (0.56) among those individuals whose age > 60 (b), who resides in urban areas and had good childhood health. While this risk shifted to 0.47 in addition to above leaf nodes among those individuals who were obese or overweight and were less than 61 years old (c). Conversely, individuals who were ST, had normal or underweight BMI and resided in rural (k) were at least risk of multimorbidity (0.19). In the same arm of BMI, a similar risk pattern (0.24) was also observed among those individuals whose age was less than 64 and who didn’t belong to the ST group (j). However, multimorbidity risk was elevated manifold for those who were more than 63 years of age (i). Level of education played a vital significant role in both single and multimorbidity risk. The risk of single morbidity was highest (0.37) among those individuals who had 10 plus years of education, had normal or underweight BMI, resided in rural and had good childhood health (d). However, this risk shrined to (0.33) for those who had no schooling or less than 5 years of schooling (e). Parallelly for multimorbidity, this risk shifted from 0.36 to 0.27.

A distinctive pattern in conditional risk shift was observed for multimorbidity among rural residents. The risk of multimorbidity is 0.53 among those Individuals whose age was greater than 56, obese or overweight, who resided in rural and who had good childhood health (f). Moreover, this risk decreases to 0.39 for those individuals whose age was less than 57 and who had education (g). However, this risk further decreases to 0.28 for those individuals who had no schooling (h).

The CART model achieved an overall classification accuracy of 88%, with a kappa statistic of 0.78, indicating substantial agreement. The model demonstrated strong sensitivity and specificity across all multimorbidity categories. A pseudo R-squared of 0.56 and a low cross-validation error indicate good model fit and reliable predictive performance.

4.1 Conclusion

In conclusion, this study for the first time incorporated thirty-six morbidities for evaluating the composition, multiplicity and examines the underlying multimorbidity susceptibility within the context of conditional risk factors among older adults in India. Nonetheless, this study also specifies the relative share of multimorbidity burden over age. The study highlighted that acute and degenerative morbidities had a huge burden across all age groups, while complex multimorbidity share expanded significantly in the oldest of old age groups. Furthermore, our results are complemented with extended evidence on the role of socioeconomic, demographic and modifiable or unmodifiable risk factors in multimorbidity composition. Childhood health was the key leading determinant of multimorbidity susceptibility and enhanced our understanding of healthy aging and place of residence, age, BMI and caste category were identified as the leading key factors for multimorbidity susceptibility among older adults in India. Finally, to address the significant issue of multimorbidity among older adults in India, it is essential to implement integrated health programs that focus on both NCDs and CDs through holistic care models in primary healthcare. Initiatives should include preventive health education targeting childhood health, socioeconomic support for vulnerable groups, and health literacy campaigns that promote healthy lifestyles.