Solid Cooking Fuel Use and Cognitive Decline among Older Mexican Adults

Abstract

Studies of air pollution and cognition often rely on measures from outdoor environments. Many individuals in low- and middle-income countries are exposed to indoor air pollution from combustion of solid cooking fuels. Little is known about how solid cooking fuel use affects cognitive decline over time. This study uses data from the 2012, 2015, and 2018 Mexican Health and Aging Study (n=14,245, age 50+) to assess how use of wood or coal for cooking fuel affects cognition of older adults relative to use of gas. It uses latent change score modeling to determine how using solid cooking fuel affected performance in Verbal Learning, Verbal Recall, Visual Scanning, and Verbal Fluency. Solid cooking fuel was used by 17% of the full sample but was more common in rural areas. Solid fuel users also had lower socioeconomic status. Compared to those using gas, solid fuel users had lower baseline scores and faster decline in Verbal Learning (β=−0.18, p<0.05), Visual Scanning (β=−1.00, p<0.001), and Verbal Fluency (β=−0.33, p<0.001). Indoor air pollution from solid cooking fuels may represent a modifiable risk for cognitive decline. Policy should focus on facilitating access to clean cooking fuels.

Introduction

Growing research has found that ambient air pollution may adversely affect cognitive outcomes among older adults ^1–6. Most of this research has focused on outdoor air pollution rather than pollutants in an individual’s home. Household air pollution, however, is highly relevant in low- and middle-income countries (LMICs) where many rely on solid fuels for domestic needs such as cooking, heating, and lighting. Such indoor use of solid fuels can expose household residents to high levels of fine particulate matter (PM_2.5)⁷ and various chemicals (e.g., carbon monoxide, polycyclic aromatic hydrocarbons, free radicals, nitrogen and sulfur oxides, mercury, lead, and magnetite) that are harmful to human health ^8–12. The World Health Organization has estimated that around three billion people worldwide rely on cooking fuels (such as kerosene, wood, and coal) which pollute indoor environments and that nearly four million premature deaths annually can be attributed to use of such fuels ¹³.

Until recently, few researchers had investigated how solid fuel use affects cognition. In 2012, researchers connected prenatal woodsmoke exposure to impairments in children’s neurodevelopment in Guatemala ¹⁴. Open-fire cooking exposure has also been associated with poorer cognitive performance among children in communities in Belize, Kenya, Nepal, and American Samoa ¹⁵. Researchers have also wondered how solid fuel use affects cognitive outcomes in mid- and late-life. Solid cooking fuel use was associated with poorer cognitive outcomes in Mexico ¹⁶ and India ¹⁷. Use of solid fuel for cooking or heating was similarly associated with poorer cognitive abilities in China ¹⁸. Open fire heating in Ireland was also associated with worse performance in multiple cognitive tasks ¹¹.

Research on solid fuel use and cognition has largely been limited to cross-sectional studies. Whether solid fuel use relates to change in cognition over time for individuals remains an open research question. Analyzing longitudinal data can better determine whether use of solid fuel predicts change in cognition. Longitudinal research can also help advance a general understanding of how air pollution affects cognitive decline. Some researchers have found high levels of ambient air pollution to predict faster cognitive decline ^19–23 although others report no relation ^24,25.

Cooking with solid fuels is more common in LMICs than elsewhere ^7,13. Although the use of solid cooking fuels has been declining in Mexico (an upper middle-income country), more than four million households there used solid cooking fuels in 2010 ²⁶. Solid cooking fuel use in Mexico is also associated with poverty and is more common in rural areas and in southern regions of the country ^26,27. Identifying modifiable risks for poor cognitive outcomes in Mexico is crucial for the aging population there ²⁸, with the number of Mexicans who live with dementia projected to increase more than four-fold between 2010 and 2050 ²⁹. This analysis seeks to understand how cognitive functioning and cognitive decline differ by whether older Mexicans cook with solid fuels or with gas in their homes.

Materials and Methods

Participants

This study uses data from the Mexican Health and Aging Study (MHAS) ³⁰. The MHAS is a longitudinal survey of Mexicans age 50 or older and their spouses regardless of age. It draws respondents across all states of Mexico and is representative of Mexico both nationally and for rural and urban areas. The MHAS survey protocols have been described in greater detail elsewhere ³¹ and a timeline of the study is available at http://www.mhasweb.org/images/Timeline_English_11_4_20.png. The MHAS has been conducted in 2001, 2003, 2012, 2015, and 2018. The MHAS was approved by the Institutional Review Boards or Ethics Committees of the University of Texas Medical Branch in the United States, the Instituto Nacional de Estadística y Geografía (INEGI) and the Instituto Nacional de Salud Pública (INSP) in Mexico. This analysis uses data from the 2012, 2015, and 2018 MHAS waves (hereafter, Times 1, 2, and 3). MHAS surveys for these waves provide cognitive information measured in regularly spaced intervals. This is ideal for the statistical approach described below ³².

Among the 15,723 MHAS respondents in 2012, I exclude from analysis the 851 respondents who were under age 50 or for whom age was not reported. Among the 14,872 remaining age-eligible respondents, I exclude the 627 individuals who were only interviewed by proxy across Times 1, 2, and 3, because proxy interviews do not include the cognitive questions I analyze. Among the 14,245 remaining respondents from the 2012 wave (95.8% of the age-eligible sample) who comprise the analytic sample, 12,601 (88.5%) were re-interviewed in 2015 and 10,717 (75.2%) were re-interviewed in 2018. As Supplemental Table 1 shows, respondents with no Time 3 (2018) cognitive data were more likely to live in urban areas, cook with gas, be male, not be married, have smoked, not have health insurance, perform worse across cognitive tasks, be older, be in a lower wealth decile, and have more chronic conditions at Time 1.

Solid Cooking Fuel Use

Both cooking with wood and cooking with coal can greatly increase household concentrations of PM_2.5 compared to homes that cook with cleaner fuels such as gas ³³. I gauge exposure to such household air pollution by assessing results from the MHAS question of respondents, “what fuel is used most for cooking?,” with response options “gas,” “wood or coal,” or “other.” I classify those who responded “wood or coal” as using solid cooking fuel and compare them to those who use gas. Ideally, this question would have separated wood and coal. However, few Mexican households (<1%) rely on coal for cooking fuel ²⁶, so the solid fuel group likely largely reflects households cooking with wood. I do not consider those who reported using “other” fuels, and I do not consider fuels used for heating and lighting, which were not asked in the 2012 MHAS. The MHAS distributions for primary household cooking fuel were similar to those of the 2010 Mexican Census ²⁶ suggesting that the MHAS accurately represents the distribution of cooking fuels in the Mexican population.

Cognitive Function:

Cognitive performance in four cognitive assessments is evaluated. Three of these (Verbal Learning, Verbal Recall, and Visual Scanning) are from the Cross-Cultural Cognitive Examination (CCCE), a battery of tests spanning several cognitive domains designed for populations with low levels of education ³⁴. The specific tasks included:

• Verbal Learning, in which the respondent is read a list of eight words and asked to immediately recall them across three trials. I use the sum of the number of words recalled across the three trials (range: 0–24).

• Verbal Recall, in which the respondent is asked to recall the eight-word list after a delay (range: 0–8). Whereas Verbal Learning captures immediate memory, Verbal Recall captures delayed memory.

• Visual Scanning, in which the respondent is given an array of visual symbols (both target and irrelevant) and asked to identify occurrences of the target symbol while ignoring irrelevant symbols (range: 0–60). This task assesses attention.

• Verbal Fluency, an MHAS task added to the three CCCE tasks above, in which respondents are asked to freely name as many animals as one can from memory for one minute (range: 0–66). This task measures executive functioning ³⁵.

The above are valid measures of cognitive outcomes in Mexico ³⁶ and have been widely used to assess cognitive function and decline in Mexico ³⁷. Analyses focus on scores across cognitive tasks at baseline as well as longitudinal change in cognitive scores (cognitive decline). The cognitive tasks were administered in Spanish and therefore required a minimum level of Spanish fluency by participants, which all respondents in the analytic sample had.

Control Variables

Solid cooking fuel use in Mexico varies by socioeconomic status (SES) and rurality, both of which are associated with cognitive function in Mexico,³⁸ making them confounding variables for analyses. Both are included as control variables.

For rurality, I categorized respondents as living in localities with 100,000+, 15,000–99,999, 2,500–14,999, or <2,500 residents). Localities with fewer than 2,500 inhabitants are recognized as “rural”, but I used this more detailed measure because both gas fuel use ^16,26 and cognitive function ³⁸ tend to be higher in more populated localities. SES measures included both years of education and household wealth (measured using the sum of the values of assets including real estate, money in accounts and stocks, vehicles, and businesses). I categorized wealth by deciles, which is treated as a continuous variable.

Chronic conditions, tobacco consumption, and health insurance access may influence rates of cognitive decline ^40,41 and differ by rural or urban residence, SES groups ^38,39, and by whether a household uses solid or gas fuel ¹⁶. These variables are included as control variables. Specific control variables were tobacco consumption (ever versus never smoker), number of chronic conditions (hypertension, diabetes, stroke, heart attack, respiratory conditions, and cancer), and whether the respondent had health insurance.

Analyses also adjusted for demographic characteristics (age, gender, and a binary variable indicating whether a respondent was married/partnered or not). I also included a binary variable indicating whether a respondent spoke an indigenous dialect to account for possible differences by language usage or familiarity with Spanish among respondents.

Statistical Analysis

The first statistical analysis used χ² and t-tests to compare characteristics of respondents who used solid cooking fuel with those who used gas. The second analysis used latent change score (LCS) models to determine how use of solid cooking fuel or gas was related to changes over time in cognitive functioning of respondents.

LCS models ^42,43 are a widely used structural equation modeling approach to understand cognitive change and between person differences in cognitive change. The LCS model estimates change in scores across measurement occasions—or, for this study, change in cognitive function between Time 1 and 2, and Time 2 and 3. The LCS model accommodates missing data on independent/dependent variables using full information maximum likelihood to reduce bias from missing data ⁴⁴. The LCS model adjusts for regression to the mean and floor/ceiling effects by including an autoregressive effect making change proportional to that at the beginning of the interval (that is, to make Time 1–2 cognitive change conditional on level of Time 1 cognition, and Time 2–3 cognitive change conditional on Time 2 cognition) ⁴⁵.

I estimate LCS models separately for each cognitive task because effects of solid fuel use may differ by cognitive abilities ^16,18. Figure 1 diagrams the LCS model. Rectangles represent observed variables, circles denote latent variables, and triangles indicate means. Key outcomes are cognition at Time 1, the Time 1–2 latent change score representing Time 1–2 cognitive change (Δ Cog. 1–2), and the Time 2–3 latent change score representing Time 2–3 cognitive change (Δ Cog. 2–3). The path from Time 1 fuel use to Time 1 Cognition (β Fuel on Cog. 1) captures differences in baseline cognition between solid fuel and gas users. Paths from fuel use to cognitive change scores (β Fuel on Δ Cog.) capture differences in cognitive change over intervals between respondents reporting using solid fuel versus gas at the start of the interval.

Figure 1.

Diagram of Latent Change Score Model

Note. “Δ Cog. 1–2” and “Δ Cog. 2–3” represent cognitive change from Time 1–2 and Time 2–3. Parameter estimate “β” represents the autoregressive effect (effect of Time 1 cognitive performance on change in cognition from Time 1–2, and effect of Time 2 cognitive performance on change in cognition from Time 2–3, which are constrained to equality). “e(cog#)” represents the residual variances of observed cognitive scores, which are constrained to equality across timepoints. Estimate “β Fuel on Δ Cog.” represents effects of solid fuel use (wood/coal versus gas) on change in cognitive scores. Estimate “β Fuel on Cog. 1” represents effects of solid fuel use (wood/coal versus gas) on baseline cognitive score. Arrows from “1” represent an estimated mean of the variable where the arrow terminates. In addition to type of cooking fuel, all control variables were included in the model and specified in a similar way to cooking fuel (paths from control to cognitive change scores and baseline cognition are estimated and all means and variances and covariances of cooking fuel and control variables are estimated as free parameters).

I treated solid fuel use as time-varying and based all other independent variables on Time 1 values. Each control variable was allowed to affect Time 1 cognition and latent change scores. I made effects of each control variable on each latent change score (Δ Cog. 1–2 and Δ Cog. 2–3) equal so as to prevent the model from being overly complex. I mean centered all independent variables and centered observed cognitive variables based on Time 1 means. Model fit was evaluated using Root Mean Square Error of Approximation (RMSEA), Comparative Fit Index (CFI), and the Tucker-Lewis Index (TLI). Fit statistics for these were within acceptable ranges for each: below 0.08 for RMSEA, above 0.90 for CFI, and above 0.90 for TLI ⁴⁷. Models were estimated with the “lavaan” R package ⁴⁸.

Results

Descriptive Results

Table 1 shows respondent characteristics at Time 1 by their use of cooking fuel. Use of solid fuel for cooking was far more common in rural than urban areas. Table 1 also shows that across all three waves respondents who used solid cooking fuel performed worse than those who used gas on all cognitive tasks. The differences between each group are statistically significant for all tests in all three waves of the survey. Respondents who relied on solid cooking fuels also had fewer years of education and less wealth than those who cooked with gas.

Table 1.

Cognitive Function and Sociodemographic Characteristics of Older Mexican Adults (Age 50+) by Reported Cooking Fuel

	Time 1 Cooking Fuel
	Gas Cooking Fuel (n=12,423, 83.3%)		Solid Cooking Fuel (n=1,753, 16.8%)
Verbal Learning					p
Time 1 (2012) (Mean SD)	14.8	3.7	12.8	3.6	***
Time 2 (2015) (Mean SD)	14.6	3.5	12.5	3.7	***
Time 3 (2018) (Mean SD)	14.7	3.8	12.4	3.7	***
Verbal Recall
Time 1 (2012) (Mean SD)	4.7	2.0	4.0	2.1	***
Time 2 (2015) (Mean SD)	4.3	2.1	3.6	2.2	***
Time 3 (2018) (Mean SD)	4.4	2.0	3.4	2.1	***
Visual Scanning
Time 1 (2012) (Mean SD)	31.9	15.4	20.1	12.4	***
Time 2 (2015) (Mean SD)	32.3	15.9	18.0	11.9	***
Time 3 (2018) (Mean SD)	32.0	15.5	17.7	11.4	***
Verbal Fluency
Time 1 (2012) (Mean SD)	15.8	5.2	12.9	4.4	***
Time 2 (2015) (Mean SD)	16.2	5.3	12.9	4.1	***
Time 3 (2018) (Mean SD)	15.9	5.3	12.6	4.1	***
Demographics (Time 1)
Age (Mean SD)	62.3	9.3	64.3	9.9
Female (n %)	7,132	53.9	919	51.2	***
Speaks Indigenous Dialect (n %)	537	3.8	462	29.9	***
Married (n %)	8,469	69.9	1,341	74.7	***
Locality Size (Time 1)
≥100,000 (n %)	8,155	57.5	123	3.8	***
15,000–99,999 (n %)	1,434	16.5	145	5.0
2,500–14,999 (n %)	1,266	13.0	287	17.1
<2,500 (n %)	1,568	13.1	1,198	74.1
Socioeconomic Markers (Time 1)
Years of Education (Mean SD)	6.5	4.9	2.4	2.7	***
Wealth Decile (Mean SD)	4.6	2.9	3.2	2.5	***
Health Variables (Time 1)
Ever Smoker (n %)	4,726	40.3	561	33.3	***
Health Insurance (n %)	10,988	86.3	1,447	80.5	***
Chronic Condition Count (Mean SD)	0.8	0.9	0.7	0.9	***

Source: Data from the 2012, 2015, and 2018 waves of the Mexican Health and Aging Study. Sample size for variables differs due to missing data. All means, standard deviations, and percentages are weighted. Column labeled “p” contains the p-values for tests of differences in cognitive function and sociodemographic characteristics between gas and solid fuel groups. P-values were calculated using t-tests for age, education, and wealth decile. They were calculated using tests of proportions for gender, indigenous dialect, marriage, smoking, and health insurance, and by using chi-square tests for locality size.

^*indicates p<0.05

^**indicates p<0.01

^***indicates p<0.001.

More than 95% of the sample had no missing data on any independent variable at baseline (cooking fuel or control variables). Nearly 83% had complete baseline data (all cognitive tasks, cooking fuel, and control variables) and 92% had no more than one missing variable at baseline. Missing data did increase on primary cooking fuel from 0.48% at Time 1 to 12.24% at Time 2, as Supplemental Table 2 shows. Missing data also increased on cognitive variables. Although 85.6% completed all cognitive tasks in Time 1, 78.1% did so in Time 2 as did 61.4% in Time 3.

Primary cooking fuel was relatively stable across time. Table 2 shows 81.8% reported using gas at all three times, whereas 8.6% reported using solid fuel across all waves. Among those who did switch, switching from solid fuel to gas was more common than switching from gas to solid. Among those who reported using gas at baseline, 97.1% still did so at Time 3, while among those who used a solid cooking fuel at baseline, 68.1% still did so at Time 3.

Table 2.

Primary Reported Cooking Fuel across Study Waves among Mexican Adults (Age 50+)

Cooking Fuel at Time 1, 2, and 3	n	%
Gas, Gas, Gas	8,385	81.76
Gas, Gas, Solid	155	1.51
Gas, Solid, Gas	202	1.97
Gas, Solid, Solid	100	0.98
Solid, Gas, Gas	263	2.56
Solid, Gas, Solid	82	0.8
Solid, Solid, Gas	192	1.87
Solid, Solid, Solid	877	8.55
Total:	10,256	100.00

Source: Data from the Mexican Health and Aging Study. Times 1, 2, and 3 represent the 2012, 2015, and 2018 study waves, respectively. “Solid” is wood/coal cooking fuel. Only respondents who had data on cooking fuel across all three time points (n=10,256) are shown, excluding 3,989 respondents for whom primary cooking fuel was missing in at least one wave. Frequencies and percentages are unweighted.

Regression Results:

Before adding predictors of baseline cognitive scores or change over time, I fit the models without predictors to understand the general pattern of cognitive change. For each cognitive task, the variances of baseline scores were significant, indicating significant cognitive differences between individuals at baseline. Cognitive abilities for all tasks but Verbal Fluency decreased between Time 1 and Time 2 and again from Time 2 to Time 3; Verbal Fluency scores increased from Time 1–2 but decreased from Time 2–3. All change scores excepting that for Verbal Recall between Times 2 and 3 had significant variances, indicating differences between individuals in rates of cognitive decline. To evaluate whether effects of solid cooking fuel use on Δ Cog. 1–2 versus Δ Cog. 2–3 differed, I tested models (with predictors) in which 1) effects were constrained to be equal and 2) effects differed. Model fit comparisons using Satorra-Bentler scaled chi-square difference tests ⁴⁶ showed no significant improvement when solid cooking fuel use effects differed across waves. I therefore constrained effects of solid fuel use on cognitive change to be equal across waves

Table 3 presents key results from LCS models, estimated separately for each cognitive outcome. There are four groups of columns labeled across the top, each for one cognitive outcome: Verbal Learning, Verbal Recall, Visual Scanning, and Verbal Fluency. The rows of the table are grouped by panels where the first panel shows effects of variables on baseline cognition and the second panel shows effects of variables on change in cognition. Whereas Table 3 only shows effects of variables on baseline cognition and cognitive change, Supplemental Table 3 contains additional panels showing means and variances of baseline cognition (essentially the model intercept at baseline, which is not discussed further), means and variances of cognitive change from Time 1–2 and Time 2–3, and the model fit statistics. Model Fit Statistics show that models for Verbal Learning, Verbal Recall, Visual Scanning, and Verbal Fluency fit well. They all have CFI and TLI scores well above 0.9, and RMSEA scores well below 0.08. I turn below to specific findings for each model.

Table 3.

Abbreviated Results from Latent Change Score Models of Cognitive Functioning among Mexican Adults age 50+ (n=14,245).

	Verbal Learning		Verbal Recall		Visual Scanning		Verbal Fluency
	β	SE	β	SE	β	SE	β	SE
Effects of Variables on Baseline Cognition
Solid Cooking Fuel	−0.52***	0.10	−0.11	0.06	−3.90***	0.33	−0.66***	0.13
Age	−0.11***	0.00	−0.07***	0.00	−0.49***	0.01	−0.11***	0.00
Education	0.24***	0.01	0.09***	0.00	1.45***	0.03	0.36***	0.01
Female	1.04***	0.06	0.58***	0.04	0.26	0.24	−0.25**	0.09
Indigenous Dialect	−0.45***	0.11	−0.05	0.06	−1.64***	0.39	−1.32***	0.15
Married	0.24***	0.06	0.05	0.04	1.07***	0.24	0.29***	0.09
Locality Size	−0.16***	0.03	−0.05***	0.02	−0.93***	0.09	−0.18***	0.04
Wealth Decile	0.03**	0.01	0.00	0.01	0.16***	0.04	0.05***	0.01
Ever Smoke	0.15*	0.06	0.02	0.03	1.27***	0.23	0.63***	0.09
Health Insurance	0.33***	0.08	0.06	0.05	0.24	0.31	0.25*	0.11
Chronic Condition Count	−0.02	0.03	−0.02	0.02	−0.70***	0.11	−0.05	0.04
Effects of Variables on Cognitive Change
Score Time(t-1)	−0.10***	0.03	−0.08*	0.03	−0.12***	0.02	−0.14***	0.03
Solid Cooking Fuel	−0.18*	0.07	−0.07	0.04	−1.00***	0.24	−0.33***	0.09
Age	−0.03***	0.00	−0.01***	0.00	−0.12***	0.01	−0.04***	0.00
Education	0.03**	0.01	0.01*	0.00	0.16***	0.04	0.05***	0.01
Female	0.18***	0.05	0.09**	0.03	0.04	0.15	−0.02	0.06
Indigenous Dialect	−0.29***	0.08	−0.03	0.04	−0.58*	0.24	−0.26*	0.10
Married	0.02	0.04	0.01	0.02	−0.02	0.15	0.04	0.06
Locality Size	−0.01	0.02	−0.02	0.01	−0.02	0.06	0.03	0.02
Wealth Decile	0.00	0.01	0.00	0.00	0.04	0.02	0.01	0.01
Ever Smoke	−0.03	0.04	0.02	0.02	0.11	0.15	−0.05	0.06
Health Insurance	0.08	0.06	0.00	0.03	0.12	0.21	0.07	0.07
Chronic Condition Count	−0.05*	0.02	−0.04**	0.01	−0.29***	0.07	−0.06*	0.03

Source: Data from the Mexican Health and Aging Study. Times 1, 2, and 3 represent the 2012, 2015, and 2018 study waves, respectively. “Solid cooking fuel” is wood/coal versus gas. β indicates unstandardized parameter estimate. SE indicates standard error. Independent variables are mean centered, whereas cognitive scores are mean centered at their Time 1 means. In addition to the parameters shown, models also estimate means and variances of baseline cognitive function and cognitive change from Time 1–2 and Time 2–3, and variances, means, and covariances of independent variables. Effects of variables on change in cognition were constrained to be equal across time. Cooking fuel is time-varying whereas all other independent variables are based on their Time 1 values. All fit statistics are within acceptable ranges.

^*indicates p<0.05

^**indicates p<0.01

^***indicates p<0.001.

Verbal Learning.

Controlling for demographic, health, and socioeconomic variables, use of solid cooking fuel rather than gas at Time 1 was associated with poorer baseline Verbal Learning. As the coefficient scores (β = −0.52, p < 0.001 in the “Effects of Variables on Baseline Cognition” panel of Table 3) show, use of solid cooking fuel was associated, on average, with a difference of more than a half-point on the Verbal Learning scale (which, as noted earlier, has a range of 0 to 24). Use of solid fuel for cooking over time was also associated with more rapid decline in Verbal Learning relative to the deterioration among those who cooked with gas. Those who reported solid cooking fuel use at the beginning of an inter-wave period had, on average, a decrease of nearly 0.2 more points in their Verbal Learning ability scores between waves relative to those who used gas for cooking (β = −0.18, p < 0.05, in the “Effects of Variables on Cognitive Change” panel).

Verbal Recall.

Differences in baseline Verbal Recall scores between solid fuel and gas users were in the expected direction but only marginally significant (β = −0.11, p < 0.10 in the “Effects of Variables on Baseline Cognition” panel). Similarly, over time, the association between solid cooking fuel use (versus gas) and cognitive decline was also in the expected direction but only marginally significant (β = −0.07, p < 0.10 in the “Effects of Variables on Cognitive Change” panel). Put another way, models indicate no more than marginal evidence that solid cooking fuel affects Verbal Recall.

Visual Scanning.

Solid fuel use does appear to affect Visual Scanning abilities. At baseline, those who used solid fuel scored, on average, nearly four points lower (on a scale ranging from 0 to 60) on the MHAS Visual Scanning test (β = −3.90, p < 0.001, as shown in “Effects of Variables on Baseline Cognition” panel). Furthermore, those who cooked with solid fuel at the beginning of a time period saw faster decline in their Visual Scanning than those who cooked with gas: on average, a decrease of 1.00 more point between waves (β = −1.00, p < 0.001, as shown in the “Effects of Variables on Cognitive Change” panel) relative to those who cooked with gas, even after accounting for all covariates.

Verbal Fluency.

Solid fuel use was also associated with lower baseline scores and more rapid decline in Verbal Fluency scores, even after accounting for all covariates, relative to those who cooked with gas. At baseline, those who cooked with solid fuel had Verbal Fluency scores of two-thirds of a point lower (β = −0.66, p < 0.001, as shown in the “Effects of Variables on Baseline Cognition” panel and on a scale of 0 to 66). Over time, those who cooked with solid fuel saw their Verbal Fluency scores decrease an additional one-third of a point between waves relative to those who cooked with gas (β = −0.33, p < 0.001, as shown in the “Effects of Variables on Cognitive Change” panel).

To make the size of relationships between solid fuel use and poorer cognition clear, I calculated effects of solid fuel use in terms of additional years of age. I did so by dividing the parameter estimates of solid fuel use on baseline/change in cognition by the respective parameter estimates for age in Table 3. For example, for Verbal Learning, the coefficient in the “Effects of Variables on Baseline Cognition” is −0.52 coefficient, whereas that for Age is −0.11. Dividing −0.52 by −0.11, I calculate that, at baseline, those who cooked with solid fuel had Verbal Learning abilities equivalent to somebody 4.7 years older than those who cooked with gas. Similarly, those who cooked with solid fuel had Visual Scanning scores equivalent to somebody 8.0 years older than those who cooked with gas. They also had Verbal Fluency scores equivalent to somebody 6.0 years older than those who cooked with gas. Calculations for differences in rates of cognitive decline between solid fuel and gas users produced similar results. Effects of solid fuel use on cognitive decline were equivalent to differences in rates of cognitive decline one would expect between individuals separated by 6.0 (Verbal Learning), 8.3 (Visual Scanning), and 8.3 (Verbal Fluency) years of age.

Sensitivity Analyses

LCS models using only complete cases were tested (Supplemental Table 4), which had largely similar results. Solid fuel use predicted lower baseline performance on all tasks except Verbal Recall and predicted faster decline in Visual Scanning and Verbal Fluency. To evaluate whether associations between solid fuel use and cognition were present across demographic groups, I re-estimated LCS models separately by gender, education (1–5 versus 6+ years), and speaking an indigenous dialect (Supplemental Table 5). Significant effects of solid fuel use on baseline cognition and cognitive decline were observed for both gender and education groups. Among men, solid fuel use was associated with poorer baseline performance on all cognitive tasks. It also led to more rapid decline in Verbal Learning and Visual Scanning and marginally more rapid decline in Verbal Fluency (p=0.062). Among women, solid fuel use was associated with worse baseline performance on all tasks except Verbal Recall and predicted more rapid decline in all tasks except Verbal Learning. Among those with 1–5 years of education, solid fuel use was associated with worse baseline performance and quicker decline in all tasks except Verbal Recall. Among those with 6+ years of education, solid fuel use was associated with lower baseline performance on all tasks except Verbal Recall and predicted faster decline in Visual Scanning and marginally faster decline in Verbal Fluency. Solid fuel use was associated with lower baseline scores for all tasks among those who did not speak an indigenous dialect and lower baseline Visual Scanning and Verbal Fluency performance for those who did speak an indigenous dialect. It was also associated with faster decline in Visual Scanning and Verbal Fluency among those who did not speak an indigenous dialect, but it was not a significant predictor of cognitive decline among those who spoke an indigenous dialect. The lack of a statistically significant relationship between solid fuel use and cognitive decline among those who spoke an indigenous dialect may reflect the smaller sample size and wider confidence intervals in the group.

Discussion

These findings are consistent with research highlighting the negative associations between ambient air pollution and cognitive function ^{1–6,19–22} and on the effects of indoor air pollution from solid fuel use on cognitive health ^11,15–18. Results suggest that exposure to indoor air pollution from solid fuel use may lead to faster decline in Verbal Learning, Visual Scanning, and Verbal Fluency. These associations were equivalent to several additional years of age (i.e., they led to additional cognitive aging), and the effects were not explained by other respondent characteristics that were analyzed.

More rapid decline in cognitive ability among individuals with greater exposure to air pollution has been reported in studies of outdoor ambient air pollution ^19–23. The mechanisms underlying the effects of air pollution on cognition are not fully understood. Fine particulate matter can reach the brain through the olfactory bulb ^49,50 and cause neuroinflammation ^51–53. Outdoor air pollution is associated with expression of inflammatory mediators in the frontal cortex and hippocampus ⁵⁴ and pro-inflammatory cytokines in the hippocampus ⁵⁵ as well as neuroinflammation in the olfactory bulb and the hippocampus ⁵⁶. Individuals with higher exposure to outdoor air pollution also have smaller white matter volumes ⁵⁷, total cerebral volume ⁵⁸, and decreased cortical thickness and gray matter volume ⁵⁹.

Few studies have explored mechanisms underlying indoor air pollution from solid cooking fuels. Hence, it is not known whether the findings on outdoor air pollution and neuroinflammation and structural brain differences generalize to indoor air pollution from solid fuels. However, woodsmoke exposure may lead to epigenetic modifications promoting inflammation and oxidative stress and pathology in the brain ⁶⁰. Coal, charcoal, and, especially, wood combustion may also elevate ambient black carbon concentrations ^61,62 and direct emissions from coal have shown high concentrations of metals, including lead ⁶³. Because black carbon ⁶⁴ and lead ⁶⁵ may negatively impact cognition, these may be plausible mechanisms connecting solid fuel use to cognition. Future studies should measure concentrations of these pollutants in homes using specific solid fuels to better understand how each may affect cognitive ability.

Solid fuel use was associated with faster decline in Verbal Learning, Visual Scanning, and Verbal Fluency, but only marginally associated with faster decline in Verbal Recall in the full sample. Understanding biological mechanisms for these differences is challenging because no study of solid fuel users has conducted both cognitive testing and neuroimaging. These findings mirror work on outdoor air pollution showing that PM_2.5 levels may increase decline in immediate recall but not delayed recall ⁶⁶. The authors speculated this may be explained by PM_2.5 differentially affecting hippocampal regions associated with encoding and retrieval. Although this may explain why Verbal Recall did not decline more rapidly among solid fuel users in the full sample, future studies of solid fuel use should include neuroimaging and cognitive testing to understand the biological pathways through which solid fuel use affects the brain and different cognitive abilities.

Findings on the lack of a significant relationship between solid fuel use and baseline Verbal Recall in the full sample are consistent with some studies ^18,38 but not others. Some previous research suggests open-fire heating may lead to diminished Verbal Recall among older adults ¹¹ and that prenatal woodsmoke exposure may lead to neurodevelopmental impairment reducing long-term figure recall of children ¹⁴. Differences in Verbal Recall findings from those of Maher et al. (2021) may be attributed to this study focusing on cooking rather than heating fuels. Differences in results with those of Dix-Cooper et al. (2012) may be attributed to differences in respondent ages and in measures of memory and solid fuel use (self-reported fuels versus carbon monoxide as a marker of woodsmoke exposure ¹⁴). Differing findings in children and older adults also points to the need for research on solid fuel use and cognitive development in childhood and its effect on late-life decline.

This analysis has limitations. First, I used primary cooking fuel as a proxy for indoor air pollution, but this alone cannot capture dose response relationships between indoor air pollution and cognition. Future studies should directly measure concentrations of specific pollutants in the home, evaluate frequency of solid cooking fuel use, ascertain what fuels are used for other purposes, such as heating and lighting, and collect detailed information on when in their life-course individuals used specific solid fuels. All this could help portray a more complete picture of individuals’ potential exposure to indoor air pollution from solid fuels. Second, outcomes were based only on cognitive performance as the MHAS does not conduct neuroimaging. Cognitive performance testing may suffer more than brain pathology assessed through neuroimaging from measurement error and bias. The lack of neuroimaging also limits one’s ability to understand how indoor air pollution may affect specific regions of the brain and cognitive processes. Third, the MHAS has substantial attrition of respondents over time. Although I used full information maximum likelihood estimation to reduce biases associated with missing data ⁴⁴, differential attrition across groups may result in biased parameter estimates. Future work should seek to further minimize attrition and evaluate how much selective attrition may influence findings. Fourth, although models were adjusted for many demographic, economic, and social characteristics, residual confounding may still be present. Studies using causal-inference methods such as instrumental variables may be an essential next step in this research.

Despite these limitations, this analysis has substantial strengths. It was based on a large, longitudinal, nationally representative sample of older adults in a country where reliance on solid cooking fuels remains prevalent, particularly in rural areas ²⁶. Longitudinal cognitive data among these populations has been scarce until recent years. The MHAS also collected many measures of socioeconomic status to account for key confounders when estimating associations between solid cooking fuel use and cognitive outcomes.

Conclusions:

This analysis suggests that public health efforts in Mexico should continue to reduce exposure to indoor air pollution by providing clean cookstoves and by increasing access to gas fuels for cooking and other domestic needs ⁶⁷. Future research should use long-term adherence to interventions as an outcome because adherence to related interventions has been low ⁶⁸. Because use of solid fuels for cooking has been more prevalent among rural dwellers and those with lower socioeconomic status, programs to reduce solid fuel use for cooking should focus on these populations and the barriers they may face in adopting gas for cooking.

Understanding decline in cognitive function is essential to understanding the course of Alzheimer’s Disease ^69,70. There are considerable differences in rates of cognitive decline among different persons ⁷¹; these differences may be influenced by individual differences in personal characteristics and exposures. This analysis suggests that use of solid fuels for cooking may lead to indoor air pollution that puts certain individuals in LMICs at risk for faster cognitive decline. Use of solid fuel for cooking is a modifiable risk. Identifying modifiable risks for Alzheimer’s disease in LMICs is critical to mitigating its social costs, particularly as their populations age and their proportion of older adults grows ⁷². Population aging in LMICs will undoubtedly be accompanied by increases in the number of people living with dementia, which is expected to increase from 27 million in 2015 to 89 million in 2050, when around two-thirds of people with dementia will be living in such countries ⁷³. Addressing the unique environmental hazards faced by individuals in LMICs, particularly those such as solid cooking fuels that may cause indoor air pollution and accelerate cognitive decline, may help improve cognitive outcomes among the current and future cohorts of older adults around the world.

Practical Implications:

Globally, solid cooking fuel use remains an important source of indoor air pollution exposure. Little is known about how indoor air pollution from solid cooking fuels relates to decline in cognitive functioning. Using a nationally representative sample of adults age 50+ in Mexico, this study finds that individuals living in households that relied on solid cooking fuel experienced faster cognitive decline relative to counterparts cooking with gas. Policy efforts should focus on reducing indoor air pollution exposure and shifting to cleaner cooking fuels.

Data Availability Statement:

The data that support the findings of this study are openly available at http://www.mhasweb.org, reference number 30.