A Quasi-Experimental Analysis of Maternal Altitude Exposure and Infant Birth Weight

Abstract

Objectives. We analyzed singleton births to determine the relationship between birth weight and altitude exposure.

Methods. We analyzed 715 213 singleton births across 74 counties from the western states of Arizona, California, Colorado, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, and Washington from January 1, 2000, to December 31, 2000. Birth data were obtained from the Division of Vital Statistics, National Center for Health Statistics, for registered births.

Results. Regression analyses supported previous research by showing that a 1000-meter increase in maternal altitude exposure in pregnancy was associated with a 75.9-gram reduction in birth weight (95% confidence interval = −84.1, −67.6). Quantile regression models indicated significant and near-uniform depressant effects from altitude exposure across the conditional distribution of birth weight. Bivariate sample-selection models showed that a 1000-meter increase in altitude exposure, over and above baseline residential altitude, decreased birth weight by an additional 58.8 grams (95% confidence interval = −98.4, −19.2).

Conclusions. Because of calculable health care–related costs associated with lower birth weight, our reported results might be of interest to clinicians practicing at higher altitudes.

Research shows that infant birth weight appears to significantly decline at higher altitude.¹ At least 2 biologically plausible explanations underlie the birth weight and altitude association, assuming the association is causal. One explanation is that maternal oxygen deprivation is involved.^2–4 Because the driving pressure for gas exchange in lungs decreases at higher altitude, lower oxygen per breath of air by pregnant mothers may induce growth-limiting hypobaric hypoxia. Others question the hypoxia pathway, noting that sufficient oxygen delivery to the uterus is achieved by hematologic and metabolic adjustments in high-altitude conditions.^5,6 More recently, researchers noted the possibility of a second biological mechanism that involves a glucose pathway. Zamudio et al⁶ found that umbilical venous and arterial glucose concentrations were lower at high altitude, resulting in lower glucose delivery to and consumption by the fetus. Anaerobic consumption of glucose by the placenta at high altitudes appears to reduce glucose availability to the fetus. Hypoglycemia may therefore also explain lower birth weights at higher altitudes.^5,6

Whatever the precise biological mechanisms involved, empirical studies reporting a negative association between altitude and infant birth weight^1,5,7–9 are questionable on 3 grounds. First, previous studies of birth weight typically analyzed smaller populations at only a handful of conveniently selected altitude locations.^9–12 With a few exceptions,^13–16 the depressant effects of altitude across a wide range of altitude positions and over a large number of birth events are unknown. Technically, many existing studies have excessive off-support inference where altitude effect estimates are pure extrapolation or off-support of the data.¹⁷ Second, it is unclear whether the depressant effect of altitude exposure operates uniformly or differently across the conditional distribution of infant birth weight.¹⁶ Third, and perhaps most importantly, households are known to self-select into higher altitude locations. In the United States, topographic variation, mountain vistas, and higher altitude settings are known amenity pull factors in migration decisions by households.^18–20 Because of residential self-selection, maternal altitude exposure is a nonrandomly assigned treatment effect.

In this study, we addressed these 3 research design issues and arrived at a more precise quantification of the depressant effects of altitude on fetal birth weight. We did so by analyzing data on 715 213 registered births across 74 counties in the Western United States in 2000. First, infant birth weight was estimated with least-squares regression to determine mean effects of altitude. We then used nonparametric regression to determine quantile effects of altitude across the conditional distribution of birth weight. Third, to causally identify the depressant effects of altitude on birth weight, we exploited information on random altitude movements of mothers during pregnancy. We present evidence showing that up-moving and down-moving mothers were statistically similar with respect to demographic characteristic composition and conditional likelihoods of both moving and up-moving in altitude during pregnancy. We used an econometric selection model to account statistically for possible correlation between mobile mothers and underlying fetal health. We conclude the article with a brief consideration of the implications of results for clinical practice at a high altitude, and how the observed depressant effects of altitude might shed light on the widely reported recent decline in birth weight in the United States.

METHODS

We obtained birth data from the Division of Vital Statistics, National Center for Health Statistics, for registered births in the western states of Arizona, California, Colorado, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, and Washington from January 1, 2000, to December 31, 2000. A total of 74 counties were observed. Mothers who gave birth in counties with populations of less than 100 000 were spatially deidentified in the National Center for Health Statistics data. Following convention, we restricted our analysis to singleton births, birth mothers ages 15 to 49 years, and infants with plausible gestational ages (more than 20 weeks). Information on maternal residence and location of birth occurrence at the county scale were used to spatially match altitude information. The western states examined have both meaningful within and between-state variation in altitude, ranging from 5 meters below sea level to 1940 meters above sea level.

Birth records contained information on variables known to influence birth weight outcomes.^21–25 Five demographic characteristic control variables were analyzed: maternal age, education, marital status, maternal race/ethnicity, and infant gender. Maternal age was divided into 3 categories: age 15 to 19, 20 to 34, and 35 to 49 years. Maternal education was categorized as less than high school (11 years of reported education or less), high school (12 years of education), some college (13–15 years of education), and college educated (16 years of education or more). Maternal race/ethnicity was categorized as non-Hispanic White, non-Hispanic Black, non-Hispanic other race, and Hispanic. Two gestation variables were analyzed: history of preterm or small-for-gestational-age infant births (1 = yes, 0 = no) and adequacy of prenatal care (1 = adequate, 0 = inadequate). Adequacy of care was determined by the Kessner Index,^25a a measure of timing and quantity of prenatal visits adjusted for gestational length. (Table A summarizes birth weight by demographic characteristics and examined gestation variables and is available as a supplement to the online version of this article at http://www.ajph.org.)

Elevation data were from the National Oceanic and Atmospheric Administration National Geophysical Data Center (http://vdatum.noaa.gov). The National Oceanic and Atmospheric Administration's Vertical Datum Digital Elevation Model files provide high resolution 1/9 arc second raster data. Cell sizes are 3 × 3 meters. Each census tract (centroid) was assigned the altitude point value (in meters) of the underlying raster cell. County altitude of maternal residence and place of birth occurrence were determined by averaging across census tract altitude values within county. In addition to measuring altitude as a continuous variable, and to test for potential nonlinear effects, we divided the distribution of altitude at the place of birth occurrence into 4 categories: less than 1000 feet (< 304.8 meters); 1000 to 2499 feet (304.8–761.7 meters); 2500 to 3999 feet (762–1218.9 meters); and 4000 feet or greater (> 1219.2 meters).

Analytic Plan

We analyzed birth weight as a continuous variable (in grams). In the first part of our analysis, we estimated infant birth weight with least-squares regression to determine mean effects of altitude, and we used nonparametric regression to determine quantile effects of altitude across the conditional distribution of birth weight. Letting W_io denote birth weight for infant i in county o, our regression estimator was modeled as

where α_i is a constant; A_io is the altitude (in meters) of the mother’s county of birth occurrence; M_i is a vector of maternal variables; C_i is a vector of child characteristics; H_i is a vector of obstetric and prenatal health care received by a mother during pregnancy; Z_i is a vector consisting of monthly dummies corresponding to the birth month of infant i to control for seasonality; S_i is a vector consisting of dummies corresponding to the state in which the birth occurred; and ɛ_io is the error term .

Following Wehby et al,¹⁶ we employed a quantile regression procedure to estimate effects of altitude at multiple points in the conditional distribution of infant birth weight.^26,27 We modeled quantile points of 0.2, 0.4, 0.6, and 0.8 in the conditional distribution of birth weight as a function of the altitude of the mother’s county of birth occurrence.

Although the preceding analyses addressed the issues of off-support inference and considered how the altitude effect might operate differently across the distribution of birth weight, maternal altitude exposure was likely a nonrandomly assigned treatment effect. Higher altitude locations are known to pull natural amenity-seeking migrants^18–20 who are statistically dissimilar to nonamenity seeking populations. To arrive at an unbiased estimate of the birth weight and altitude relationship, one must account for risk factors of low birth weight that are correlated with the decision to reside at higher altitude. However, to the extent that risk factors of low birth weight are unrelated to a quantified change in altitude, whether up or down, the bias of residential selection can be minimized. As shown in the following, we exploited a circumstance that appeared to satisfy this analytical condition involving random altitude movement of mothers during pregnancy.

Approximately 71 200 mothers birthed infants in locales different from their reported county of residence. Taking the altitude difference of the place of birth occurrence (A_io) and place of residence (A_ir), we derived whether a mother moved up or down in altitude to birth her child and the precise change in elevation.

Although the distance moved from place of residence to place of birth occurrence was highly positively skewed (S = 17.07; i.e., moving mothers were substantially more likely to travel shorter than longer distances), the distribution of altitude movement during pregnancy had near zero skew (S = −0.27). Moreover, on the known risk factors of low birth weight, up- and down-moving mothers were statistically similar. As shown in Table 1, up and down altitude-moving mothers were nearly identical in age structure, racial composition, educational attainment, marital status, adequacy of prenatal care received, and proportion carrying to full term (> 37 weeks).

TABLE 1—

Descriptive Statistics for Up and Down Altitude Moving Mothers: Division of Vital Statistics, National Center for Health Statistics; 10 US States; 2000

Variable	Up Altitude, Proportion (SD)	Down Altitude, Proportion (SD)	Nonmovers, Proportion (SD)
Maternal age, y
15–19	0.08 (0.27)	0.08 (0.27)	0.11 (0.31)
20–34	0.75 (0.43)	0.75 (0.43)	0.75 (0.44)
35–49	0.17 (0.37)	0.18 (0.38)	0.14 (0.35)
Maternal education
< high school	0.18 (0.39)	0.16 (0.37)	0.28 (0.45)
High school	0.28 (0.45)	0.28 (0.45)	0.29 (0.46)
Some college	0.23 (0.42)	0.25 (0.43)	0.21 (0.41)
College	0.30 (0.46)	0.31 (0.46)	0.21 (0.41)
Maternal race/ethnicity
White	0.55 (0.50)	0.57 (0.50)	0.41 (0.49)
Black	0.05 (0.21)	0.05 (0.22)	0.06 (0.23)
Other	0.12 (0.32)	0.11 (0.31)	0.10 (0.30)
Hispanic	0.29 (0.45)	0.27 (0.44)	0.43 (0.49)
Maternal marital status
Married	0.76 (0.43)	0.77 (0.42)	0.67 (0.47)
Nonmarried	0.24 (0.43)	0.23 (0.42)	0.33 (0.47)
Infant sex
Male	0.51 (0.50)	0.51 (0.50)	0.51 (0.50)
Female	0.49 (0.50)	0.49 (0.50)	0.49 (0.50)
Gestation length
Full term (≥ 37 wk)	0.18 (0.38)	0.18 (0.38)	0.16 (0.37)
Premature (< 37 wk)	0.82 (0.38)	0.82 (0.38)	0.84 (0.37)
Prior preterm birth
Yes	0.01 (0.10)	0.02 (0.10)	0.01 (0.10)
No	0.99 (0.10)	0.98 (0.10)	0.99 (0.10)
Altitude
Δ altitude (1000 m)	0.10 (0.16)	−0.13 (0.18)
Distance (km)	93.7 (325.6)	65.9 (155.9)
Occurrence altitude
< 1000 ft (low)	0.60 (0.49)	0.71 (0.46)	0.67 (0.47)
1000–2499 ft	0.12 (0.33)	0.06 (0.24)	0.18 (0.39)
2500–3999 ft	0.01 (0.11)	0.00 (0.02)	0.03 (0.16)
≥ 4000 ft (high)	0.27 (0.44)	0.23 (0.42)	0.12 (0.32)
Likelihood of moving
Mean	0.111 (0.036)	0.113 (0.037)
Q1	0.055 (0.006)	0.055 (0.006)
Q2	0.074 (0.006)	0.074 (0.006)
Q3	0.098 (0.008)	0.097 (0.008)
Q4	0.123 (0.006)	0.123 (0.006)
Q5	0.153 (0.022)	0.154 (0.023)
Likelihood of up-moving
Mean	0.374 (0.034)	0.370 (0.033)
Q1	0.337 (0.026)	0.334 (0.029)
Q2	0.361 (0.004)	0.361 (0.004)
Q3	0.337 (0.026)	0.334 (0.029)
Q4	0.361 (0.004)	0.361 (0.004)
Q5	0.371 (0.004)	0.371 (0.004)

Note. Q = quantile. The 10 US states were AZ, CA, CO, ID, MT, NV, NM, OR, UT, and WA.

To support the notion that up versus down movement in altitude was a near-random process, we first compared up-moving and down-moving mothers on the conditional likelihood of changing altitude position in pregnancy and then compared up-moving and down-moving mothers on the conditional likelihood of up-moving in altitude. We regressed mover status (1 = moved during pregnancy; 0 = did not move during pregnancy) on variables influencing the decision to move (see data available as a supplement to the online version of this article at http://www.ajph.org). Up-moving and down-moving mothers were then compared on the likelihood of moving at the mean and across quintiles (reflecting varying levels on the likelihood to move). The same comparisons were made with respect to the likelihood of up-moving in altitude. Limiting to movers only, we regressed up-moving status (1 = up-moving mother; 0 = down-moving mother) on relevant covariates and then compared up-moving and down-moving mothers on the predicted probability of up-moving at the mean and at quintiles in the likelihood distribution of up-moving. If altitude movement was a sufficiently random process, allowing us to address the bias of residential selection, then we would have seen no meaningful differences between up-moving and down-moving mothers on the likelihood of moving and on the likelihood of moving upward in altitude. In addition to demographic characteristics, we reported predicted probabilities (Table 1) showing that up-moving and down-moving mothers were strikingly similar on the likelihood of moving in pregnancy, and that down-moving mothers were similar to up-moving mothers on the likelihood of up-moving during gestation.

Incorporating maternal altitude movement into our model, birth weight (W_i) for infant i in county o was then modeled as

where α_i is a constant; A_ir is the altitude (in meters) of the mother’s county of residence; G_i indicates a birth outside the mother’s county of reported residence; ΔA_io = A_io − A_ir indicates the altitude difference between the place of birth occurrence (A_io) and place of residence (A_ir); and D_io denotes the great-circle distance (in kilometers) between place and residence and birth occurrence traveled by a mother during pregnancy. Distance was calculated as , where φ₁ is the centroid of the county of birth occurrence; φ₂ is the centroid of the county of residence; and Δλ is the absolute difference in centroid between county of birth occurrence and residence. ΔA_io = 0 and D_io = 0 indicate nonmovers. The remaining components carry over from equation 2. As with previous research, we expected baseline altitude, A_ir, to be negatively associated with birth weight. Adjusting for baseline residential altitude, ΔA_io represented a pseudo-randomly assigned exposure to altitude among mothers birthing outside their county of residence, with β₃ representing our dose–response coefficient that we expected to behave negatively.

The model in equation 2 represents a quasi-experimental estimator that treated mothers who resided and birthed in the same location as control participants and mothers who did not as experimental participants. Despite adjustments for known risk factors for low birth weight, estimated effects of altitude on birth weight might be biased if unobserved heterogeneity between control and experimental mothers existed and was correlated with either altitude or birth weight. On the known risk factors of low birth weight, moving mothers were statistically dissimilar to sedentary (i.e., nonmoving) mothers.

As shown in Table 1, sedentary mothers were more likely to be younger (15–19 years of age), of Hispanic ethnicity, unmarried, and significantly less likely to possess a college degree. Thus, selection might have biased the estimated effect of change in altitude on birth weight if moving was either indicated by complications associated with lower birth weight or if birth weight was influenced by characteristics of mothers associated with a move during pregnancy. Therefore, we also estimated a bivariate sample-selection model,²⁸ accounting for the decision of a mother to move during pregnancy. In this model, we focused on birth weight of infants born to mothers who birthed in a location other than their county of residence, but we also used information provided by all mothers to account for the probability of sample selection.

A mother’s latent propensity to birth in nonresidence counties () was modeled in reduced form as

where α_1i is a constant; covariate vectors M_i, H_i, and Z_i are as described previously; S_ir is a vector for state of residence; P_i represents various pregnancy-related complications or congenital defects observed in birth records; and ε_oi is an independent normally distributed error term. A birth occurred in a county outside a mother’s county of residence (O_i = 1) when , otherwise O_i = 0. Letting W_i denote birth weight for infant i whose birth occurred in county o and mother resided in county , birth weight was modeled in reduced form as

where α_2i is a constant; A_ir, ΔA_io, D_io and vectors M_i, C_i, H_i, Z_i and S_io are the same as in equations 1 and 3; and ε_wi is an independent normally distributed error term. Both outcomes were confirmed to be independently normally distributed, and the bivariate sample selection model assumed ε_wi to be distributed bivariate normal with ε_oi (). All parameters of the joint distribution were estimated simultaneously using maximum likelihood estimation.

To ensure robust results for equation 3, we included exclusion restrictions in equation 4 using various identifiers for pregnancy-related complications in P_i that might predict the likelihood of a mother birthing in a county outside her place of residence. For instance, a mother experiencing pregnancy-related complication might seek specialized care at a distant facility with adequate resources. Pregnancy-related complications included premature rupture of membrane, abruption placenta, placenta previa, dysfunctional labor, breech presentation, cord prolapsed, maternal cardiac disease, maternal genital herpes, hydramnios, hemoglobinopathy, eclampsia, incompetent cervix, pregnancy-related hypertension, chronic hypertension, renal disease, and Rhesus sensitization.

RESULTS

Mean altitudes and birth weights for the 74 observed counties of birth occurrence were calculated (Table B, available as a supplement to the online version of this paper at http://www.ajph.org). Figure 1 summarizes this information as a scatterplot, with mean county altitude on the x-axis and mean birth weight on the y-axis. Counties are distinguished by state location, and a best-fit linear solution of the association intersects the space. The plot shows that mean county birth weight declined in mean county altitude (R² = 0.59). More precisely, we found that a 1000-meter increase in altitude decreased average birth weight by 97.4 grams (95% confidence interval [CI] = −166.7, −78.2 g). This aggregate result of approximately 100 grams of loss in 1000 meters of altitude was consistent with previous research.^1,7

FIGURE 1—

Scatterplot of average birth weight (in grams) and altitude (in meters) in county of occurrence: Division of Vital Statistics, National Center for Health Statistics; 10 US states; 2000.

Least Squares and Quantile Regression Results

Table 2 reports parameter estimates predicting birth weight. Column 1 reports the mean effect of altitude on birth weight for all singleton births. We found that a 1000-meter increase in altitude in the county of birth occurrence decreased birth weight by 71.34 grams (95% CI = −79.6, −63.1). Column 2 reports results where the distribution of county altitude was divided into 4 discrete categories, checking whether birth weight might decline nonlinearly with altitude. We found that infants born at the altitude level of 1000 to 2499 feet (304.8–761.7 meters) were 18.41 grams smaller (95% CI = −22.8, −14.0) than children born at the altitude level of less than 1000 feet (< 304.8 meters). At 2500 to 3999 feet (762–1218.9 meters), infants were 51.61 grams smaller (95% CI = −62.3, −41.0), and infants born at 4000 feet or greater (> 1219.2 meters) were on average 80.84 grams smaller (95% CI = −95.3, −66.4) than infants delivered at altitudes less than 1000 feet. Columns 3 and 4 restrict analysis to full-term infants (> 37 weeks). Results behaved similarly for full-term infant models across altitude levels, suggesting that the birth weight effect of altitude exposure likely functioned to restrict intrauterine growth.

TABLE 2—

Results From Ordinary Least Squares Regression of Birth Weight (in Grams) for All and Full-Term (> 37 Weeks) Singleton Births: Division of Vital Statistics, National Center for Health Statistics; 10 US States; 2000

Variable	Model 1: All b (95% CI)	Model 2: All b (95% CI)	Model 3: Term b (95% CI)	Model 4: Term b (95% CI)
Altitudea (m/1000)	−71.34* (−79.55, −63.13)		−68.91* (−76.45, −61.37)
< 1000 ft (Ref)
1000–2499 ft		−18.41* (−22.80, −14.03)		−18.84* (−22.87, −14.81)
2500–3999 ft		−51.61* (−62.28, −40.95)		−49.23* (−59.06, −39.40)
≥ 4000 ft		−80.84* (−95.30, −66.38)		−82.60* (−95.97, −69.21)
Constant	3204.8* (3196.1, 3213.4)	3195.9* (3187.0, 3204.8)	3306.5* (3298.5, 3314.4)	3298.7* (3290.5, 3306.9)
Model statistics
No.	715 213	715 213	599 721	599 721
Adjusted R2	0.043	0.043	0.054	0.053
F	964.0	908.1	1028.4	986.6

Note. CI = confidence intervals. All models control for birth month and state of birth occurrence, maternal age, educational attainment, race/ethnicity, marital status, adequacy of care, previous preterm birth, and gender of child. Model 1 reports the mean effect of altitude on birth weight for all singleton births. Model 2 reports results where the distribution of county altitude was divided into 4 discrete categories, checking whether birth weight might decline nonlinearly with altitude. Models 3 and 4 restrict analysis to full-term infants (> 37 weeks), with model 4 reporting results where the distribution of county altitude was divided into 4 discrete categories. The 10 US states were AZ, CA, CO, ID, MT, NV, NM, OR, UT, and WA.

^a1000 feet (304.8 m); 2499 feet (761.7 m); 2500–3999 feet (762-1218.9 m); 4000 feet (1219.2 m).

*P < .01. These are two-sided P values.

Table 3 reports quantile parameter estimates predicting singleton birth weight for all infants. We found that a 1000-meter increase in county altitude of birth occurrence significantly reduced infant birth weight at the 0.2 quantile by 64.41 grams (95% CI = −75.2, −53.6), at the 0.4 quantile by 63.07 grams (95% CI = −71.6, −54.5), at the 0.6 quantile by 68.43 grams (95% CI = −77.4, −59.5), and at the 0.8 quantile by 76.14 grams (95% CI = −85.9, −66.3). Quantile regression results indicated that a unit increase in the altitude of a mother’s county of birth occurrence consistently negatively (and near-uniformly) depressed infant birth weight across the conditional distribution of birth weight.

TABLE 3—

Results From Quantile Regression of Birth Weight (in Grams) for All Singleton Births: Division of Vital Statistics, National Center for Health Statistics; 10 US States; 2000

Variable	Q20, b (95% CI)	Q20, b (95% CI)	Q40, b (95% CI)	Q40, b (95% CI)	Q60, b (95% CI)	Q60, b (95% CI)	Q80, b (95% CI)	Q80, b (95% CI)
Altitudea (m/1000)	−64.41* (−75.20, −53.63)		−63.07* (−71.62, −54.51)		−68.43* (−77.37, −59.49)		−76.14* (−85.94, −66.34)
< 1000 ft (Ref)
1000–2499 ft		−16.50* (−22.14, −10.86)		−16.24* (−21.01, −11.47)		−17.95* (−23.04, −12.86)		−19.71* (−25.03, −14.40)
2500–3999 ft		−48.00* (−61.74, −34.26)		−43.72* (−55.34, −33.11)		−45.32* (−57.69, −32.95)		−55.94* (−68.81, −43.07)
≥ 4000 ft		−69.50* (-88.45, −50.55)		−74.59* (−90.40, −58.77)		−78.63* (−95.31, −61.95)		−82.88* (−99.99, −65.77)
Constant	2825.0* (2813.6, 2836.5)	2816.5* (2805.0, 2828.0)	3117.3* (3108.3, 3126.3)	3109.4* (3099.7, 3119.1)	3346.8* (3337.4, 3356.1)	3338.0* (3327.7, 3348.3)	3618.4* (3608.2, 3628.6)	3609.6* (3598.8, 3620.4)

Note. CI = confidence intervals; Q = quantiles. Total sample size was n = 715 213. All models control for birth month and state of birth occurrence, maternal age, education, race/ethnicity, marital status, adequacy of care, prior preterm birth, and sex of child. The 10 US states were AZ, CA, CO, ID, MT, NV, NM, OR, UT, and WA.

^a1000 feet (304.8 m); 2499 feet (761.7 m); 2500–3999 feet (762-1218.9 m); 4000 feet (1219.2 m).

*P < .01. These are two-sided P values.

Figure 2 extends and summarizes results from Tables 4 and 5. It plots altitude coefficients and estimated quantile regressions at quantile points τ = 0.05 to τ = 0.95. The shaded gray area around point estimates depicts 95% CIs. The gray solid line intersecting Figure 2 at −71.34 corresponds to the least-squares estimate of the conditional mean effect of altitude, and dashed lines at −79.6 and −63.1 represent 95% CIs for the least-squares estimate reported in column 1 of Table 2. Overall, results implied that the depressant effects of altitude on birth weight operated near uniformly across the conditional distribution of birth weight. Altitude exposure in pregnancy appeared to shift the distribution of birth weight leftward.

FIGURE 2—

Estimates from ordinary least squares and quantile regressions of birth weight (in grams) on altitude (in meters): Division of Vital Statistics, National Center for Health Statistics; Arizona, California, Colorado, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, and Washington; 2000.

Note. CI = confidence intervals.

TABLE 4—

Results From Ordinary Least-Squares (OLS) Regressions and Bivariate Sample-Selection Model on Birth Weight (in Grams) for All Singleton Births: Division of Vital Statistics, National Center for Health Statistics; Arizona, California, Colorado, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, and Washington; 2000

	OLS			Bivariate Selectiona
Variablea	Full Sample	Nonmovers	Movers	Full Sample
Residential altitude, coefficient (95% CI)	−75.86* (−84.11, −67.60)	−75.47* (−83.85, −67.10)	−103.1* (−145.4, −60.7)	−107.8* (−150.1, −65.5)
Δ Altitude, coefficient (95% CI)	−56.26* (−76.96, −35.57)		−63.01* (−102.7, −23.4)	−58.79* (−98.43, −19.15)
Distance, coefficient (95% CI)	−0.11* (−0.13, 0.10)		−0.10* (−0.12, −0.08)	−0.09* (−0.12, −0.07)
Nonmover (Ref)
Mover, coefficient (95% CI)	−30.17* (−34.81, −25.53)
Constant, coefficient (95% CI)	3208.0* (3199.3, 3216.6)	3210.6* (3201.7, 3219.5)	3148.4* (3108.3, 3188.6)	2515.14* (2438.6, 2591.7)
Model statistics
ρ				0.424
Σ				625.6
λ				265.2
No.	715 213	643 981	71 232	714 338
Adjusted R2	0.043	0.043	0.044
F	897.35	875.14	95.36
Log likelihood				−764 407.7
Wald χ2 (35)				3473.0
LR test for selection				220.79

Note. CI = confidence intervals; LR = likelihood ratio. All models control for birth month and state of birth occurrence, maternal age, education, race/ethnicity, marital status, adequacy of care, previous preterm birth, and gender of child.

^aSelection variables include all of the previous variables plus binary indicators for premature rupture of membrane, abruption placenta, placenta previa, dysfunctional labor, breech presentation, cord prolapsed, maternal cardiac disease, maternal genital herpes, hydramnios, hemoglobinopathy, eclampsia, incompetent cervix, pregnancy-related hypertension, chronic hypertension, renal disease, and Rhesus sensitization. (Results of the model for propensity to birth in a location outside of residential county is located as data available as a supplement to the online version of this article at http://www.ajph.org.)

*P < .01. These are two-sided P values.

Quasi-Experimental Results

Table 4 reports regression coefficients predicting birth weight as a function of the altitude of maternal place of residence (A_ir), the difference (ΔA_io) in altitude from the place of birth occurrence (A_io) and place of residence (A_ir), adjusting for the distance between place of birth occurrence and residence (D), and a suite of relevant health inputs (M, C, H, Z, S) as previously described. Analyzing all mothers, we found that a 1000-meter increase in the baseline county of residence was associated with a 75.86 gram (95% CI = −84.1, −67.6) reduction in infant birth weight. For every 1000 meters a mother actually moved up in altitude during gestation, birth weight declined by an additional 56.26 grams (95% CI = −77.0, −35.6).

Next, we estimated the same model but restricted the analysis to only mothers who did move in altitude to deliver a child (Table 4). Adjusting for baseline residential county altitude, the distance traveled between place of residence and place of birth occurrence, and relevant covariates, we found that an up-movement of 1000 meters (from baseline altitude) during pregnancy resulted in an additional 63.01 grams (95% CI = −102.7, −23.4) of loss in infant birth weight. The altitude effect on birth weight behaved approximately symmetrically in up-moving versus down-moving mothers. That is, the average amount gained (in grams) for a 1 000-meter descent in altitude (b = 110.9; P <.05) was approximately equal to the average amount of lost (in grams) by a 1000-meter ascent in altitude ( b= −63.01; P <.01).

Table 4 reports estimates from the birth weight equation in the bivariate sample-selection model. Similar to our mover-only model in column 3, we found that a 1000-meter increase in altitude at maternal residence decreased birth weight by 107.8 grams (95% CI = −150.1, −65.5). Also, for every 1000 meters a mother moved up in altitude from her place of residence to birth her child, birth weight declined by an additional 58.79 grams (95% CI = −98.43, −19.15). The distance moved during pregnancy was also associated with significant loss of birth weight, decreasing 0.09 grams (95% CI = −0.12, −0.07) for every kilometer traveled.

We also estimated the likelihood of the birth to occur outside the residence county (data available as a supplement to the online version of this article at http://www.ajph.org). A likelihood ratio test of independent equations could not reject the null hypothesis (P ≤ .001), indicating improved estimates with the bivariate sample-selection model. Most pregnancy-related complications and previous preterm birth were significantly associated with greater odds of moving. Variables associated with higher socioeconomic status (i.e., age, education, White, married) and greater adequacy of care were also associated with greater odds of moving. Thus, mothers with pregnancy-related complications and better access to care were more likely to deliver outside their residence county.

DISCUSSION

In this article, we aimed to extend existing empirical literature by first estimating the association of maternal altitude exposure and infant birth weight across a wider range of altitude positions and over a larger number of birth events, minimizing the technical issue of off-support inference characterizing a sizable fraction of previous research. Our reported results corroborated the direction of existing results—maternal altitude exposure in pregnancy is negatively associated with birth weight. First, we found substantively large depressant effects in altitude. For every 1000-meter increase in altitude, average birth weight decreased by approximately 76 grams. Second, in nonparametric regressions, and given overlapping intervals of confidence, we found that the depressant effect of altitude exposure in pregnancy behaved similarly across the conditional distribution of birth weight.

Third, we aimed to address the problem of nonrandom altitude exposure by exploiting pseudorandom maternal altitude movement in pregnancy. Although mothers residing at higher altitudes were demographically different than mothers residing at lower altitudes, the subset of mothers moving up or down altitude during pregnancy was sufficiently homogeneous to derive the causal effect of altitude of birth weight. Results from our quasi-experimental design strategy indicated that a 1000-meter increase in altitude exposure in pregnancy, over and above baseline residential altitude, decreased infant birth weight by an additional 56 to 63 grams. Combined with a baseline residential altitude effect of approximately 76 to 108 grams of loss for every 1000 meters of altitude, our results were slightly higher than the reported effect of 100 grams of loss for every 1000 meter of rise in altitude.^1,7 Our estimated effect sizes were also consistent with studies that examined populations in less developed nations where access to health services is more constrained,^8,10,12,16 suggesting a need to more fully understand the biological mechanisms involved.

From a health policy point of view, the substantively large depressant effects of altitude exposure in pregnancy might warrant consideration. There were calculable health care–related costs associated with pregnancies at high altitude, particularly with respect to small-for-gestational age children who are more likely to encounter various developmental challenges as they mature.^29,30 These costs are borne in various proportions by the family, an insurance provider, or the public sector, depending on the economic status of the household. Thus, there might be both private and broader societal benefits to providing additional health services to pregnant mothers at high altitude, or in extreme cases involving mothers with medical complications or conditions, encouraging pregnant mothers to move to lower altitudes during the gestation period.

Finally, in terms of population growth from 2000 to 2010, the Western states examined in this study–Arizona (2nd, 24.6%), California (20th, 10.0%), Colorado (9th, 16.9%), Idaho (4th, 21.2%), Montana (21st, 9.7%), Nevada (1st, 32.5%), New Mexico (15th, 13.2%), Oregon (18th, 12.0%), Utah (3rd, 23.8%), and Washington (13th, 14.1%)–rank among the fastest growing in the United States. The US population is drifting west. This fact is coupled with recent and highly touted studies reporting puzzling declines in birth weight and fetal growth independent of pregnancy length.^31,32 Adjusting for maternal and newborn characteristics, Morisaki et al.³¹ found that average birth weight among singletons over the last decade decreased 31 to 42 grams. Similarly, Donahue et al. found that infants born in 2005 were 52 grams smaller than infants in 1990, noting that:

Recent decreases in fetal growth among US, term, singleton neonates were not explained by trends in maternal and neonatal characteristics, changes in obstetric practices, or concurrent decreases in gestational length.^32(p357)

Between 1990 and 2000, the proportion of births in the contiguous United States at less than 1000 feet increased by 11.99% (from 0.135 to 0.151, respectively), by 22.38% (from 0.043 to 0.052, respectively) at less than 2500 feet, and by 26.30% (from 0.029 to 0.037, respectively) at less than  4500 feet in elevation. Given the results that showed large depressant effects of altitude on birth weight, some fraction of the widely observed decline in birth weight might be attributable to the drift upward in residential altitude of birth mothers. We will consider this possibility in future research.