Skip to main content
American Journal of Public Health logoLink to American Journal of Public Health
. 2019 Jul;109(7):1050–1052. doi: 10.2105/AJPH.2019.305015

Causes of Excess Deaths in Puerto Rico After Hurricane Maria: A Time-Series Estimation

Raul Cruz-Cano 1,, Erin L Mead 1
PMCID: PMC6603484  PMID: 30998411

Abstract

Objectives. To determine the number and causes of excess deaths in Puerto Rico after Hurricane Maria made landfall on September 20, 2017.

Methods. We obtained monthly vital statistics data on all deaths from January 2008 through October 2017. We conducted a time-series analysis to estimate excess mortality in September and October 2017 overall and by age, sex, and cause of death.

Results. We estimated a total of 1205 excess deaths (95% confidence interval [CI] = 707, 1702). Excess deaths were slightly higher among men than women (632 and 579 deaths, respectively) and found only among people aged 60 years or older (1038 deaths). Most excess deaths occurred from heart disease (253 deaths), “other” causes (204 deaths), diabetes (195 deaths), Alzheimer’s disease (122 deaths), and septicemia (81 deaths).

Conclusions. The number of excess deaths was similar to recent government estimates. However, this study is the first to identify the causes of death that were exacerbated by the disaster.

Public Health Implications. An accurate estimation of the top causes of excess mortality can help authorities plan resource allocation for the island’s recovery and for the prevention of deaths in future disasters.


Excess mortality after a crisis is difficult to estimate, and frequently used methods, such as surveys, are prone to bias.1 In the case of hurricanes, official estimates may not adequately capture deaths that are indirectly attributable to the event, such as deaths resulting from lack of medical care access.2

The effect of Hurricane Maria on the number of deaths in Puerto Rico has been controversial. In brief, the US territory has experienced significant population loss and an economic recession since 2006.3 The category 4 hurricane made landfall on September 20, 2017, and caused $90 billion in damages.4 Early studies of the excess mortality associated with the hurricane failed to account for long-term trends and seasonal patterns in mortality, as well as significant secular changes in population size.4,5

Time-series analyses have successfully integrated information on secular and seasonal trends into accurate estimates of mortality.6 In 2018, researchers in a study commissioned by the Puerto Rico government conducted a time-series analysis and estimated 1271 excess deaths from Hurricane Maria.7 However, this study did not address the causes of death exacerbated by the hurricane. Using 2008 to 2017 vital statistics data, we applied time-series modeling to quantify the excess mortality from Hurricane Maria in Puerto Rico overall and by cause, sex, and age.

METHODS

We started with 2008 data because the current life expectancy trend began in that year.8 We obtained monthly death counts for the period 2008 to 2016, including causes of death, from the Puerto Rico Vital Statistics System via e-mail. Vital Statistics System data for 2017 were available online.7 Vital Statistics System data are extracted from death certificates in the local offices of the demographic registries of most municipalities.9

Because of the significant variations in life expectancy and population size, we calculated mortality rates per 1000 residents rather than raw death counts.6 To calculate monthly population estimates, we used annual estimates from the American Community Survey 5-Year Estimates10 as the population size for January of each year. Within the range of known data points from January months, we used linear interpolation to construct monthly population estimates for the remaining months. The 5-year American Community Survey 2018 estimate was not available. Therefore, we used linear extrapolations of January 2016 and 2017 data to estimate population sizes for February through October 2017.

We used an auto-regressive integrated moving-average (ARIMA) model to analyze our data; these models incorporate past values to predict future values while accounting for seasonal patterns and time trends.11 A seasonal ARIMA model is defined as (p, d, q) × (P, D, Q)s, where p, d, and q represent the autoregressive order, number of differences, and moving average order, respectively; P, D, and Q represent the seasonal autoregressive order, number of seasonal differences, and seasonal moving average order, respectively; and s is the length of the seasonal period.

Monthly mortality rates from January 2008 to August 2017 were used as the modeling data set, and rates from September and October 2017 were the forecasting data set. From the modeling data set, the model was selected and confirmed as an adequate representation of mortality rates via autocorrelation function, partial autocorrelation function, and residual white noise probability graphs.11 Graphs showed that a nonstationary seasonal time-series model was appropriate and that seasonal and nonseasonal auto-regressive and moving-average components were present.

These observations led to a final (1, 1, 1) × (1, 1, 1)12 model (the modeling equation is shown in Appendix A, available as a supplement to the online version of this article at http://www.ajph.org). Only 3 months fell outside the 95% confidence interval (CI), which suggested model adequacy (Appendix B, available as a supplement to the online version of this article at http://www.ajph.org). To further confirm model fit, we used the ARIMA model to predict the monthly mortality rates for January to August 2017 (prior to the hurricane) and compared them with the observed mortality rates. All predicted values fell within the 95% confidence interval, indicating a good fit and accurate predictions.

We stratified our analysis by sex, age group (0–9 years, 10–19 years, 20–59 years, ≥ 60 years), and the 42 causes of death tracked and reported by the Puerto Rican Department of Health. We limited the analysis to the top 5 causes of excess deaths (heart disease, other causes, diabetes, Alzheimer’s disease, and septicemia). According to autocorrelation function, partial autocorrelation function, and residual white noise graphs, all of the models fit with the exception of septicemia, for which we fit a (2, 1, 1) × (2, 1, 1)12 ARIMA model.

Our sample size was sufficient to use continuous-outcome ARIMA models to estimate total and stratified mortality rates. Our data set of 118 observations exceeded the recommended 50 or more observations to estimate the autocorrelation function.11 None of the predictions, including the lower bounds of their 95% confidence intervals, were close to zero; hence, this assumption of continuous-outcome models was not broken by the zero bound of the mortality rates.

We calculated the differences between the observed and the expected mortality rates for September and October 2017 to estimate the number of excess deaths. Because we used nondeterministic models, the number of excess deaths from the stratified models may not precisely match the model for the total population in a likely case of a Simpson’s paradox. SAS version 9.4 (SAS Institute Inc, Cary, NC) was used in conducting all analyses.

RESULTS

Expected mortality rates were 8.08 (95% CI = 7.31, 8.87) per 1000 people in September 2017 and 8.24 (95% CI = 7.32, 9.16) per 1000 in October 2017 (Table 1 and Appendix B). Observed mortality rates were 10.20 per 1000 in September and 10.26 per 1000 in October. In total, there were 1205 excess deaths (95% CI = 707, 1702): 607 (95% CI = 381, 832) in September and 598 (95% CI = 326, 870) in October. Appendix C (available as a supplement to the online version of this article at http://www.ajph.org) compares our estimates with those of other reports.

TABLE 1—

ARIMA Model Forecasts of Mortality Rates Overall and by Sex, Age, and Top Five Causes of Death: Puerto Rico, September and October 2017

Variable and Month Population Estimate, No. Observed Ratea Forecast Ratea (95% CI) Estimated Deaths, No. Excess Deaths, Point Estimate (SE; 95% CI)
Total excess deaths
 September 3 493 593 10.20 8.08 (7.30, 8.87) 2 321.48 607 (114.85; 381, 832)
 October 3 489 119 10.26 8.24 (7.32, 9.16) 2 442.12 598 (138.84; 326, 870)
Men
 September 1 667 072 11.68 9.09 (8.16, 10.02) 2 609.73 356 (136.60; 228, 483)
 October 1 664 732 11.34 9.39 (8.31, 10.46) 2 781.41 276 (162.41; 124, 428)
Women
 September 1 826 521 8.84 7.13 (6.29, 7.97) 2 048.20 256 (133.91; 130, 382)
 October 1 824 387 9.27 7.19 (6.31, 8.08) 2 130.90 323 (136.11; 186, 460)
Aged ≥ 60 yb
 September 840 835 35.38 27.60 (24.34, 30.86) 7 926.03 537 (477.40; 312, 763)
 October 842 230 35.35 28.36 (24.59, 32.12) 8 402.87 501 (569.83; 231, 770)
Heart disease
 September . . . 1.95 1.56 (1.33, 1.79) 448.63 105 (33.35; 43, 166)
 October . . . 2.10 1.57 (1.31, 1.82) 463.86 148 (38.85; 76, 219)
Other causesc
 September . . . 1.83 1.42 (1.18, 1.65) 406.51 111 (34.93; 47, 176)
 October . . . 1.79 1.45 (1.19, 1.72) 431.00 93 (40.68; 18, 168)
Diabetes
 September . . . 1.30 0.88 (0.68, 1.07) 251.28 115 (28.09; 63, 166)
 October . . . 1.13 0.84 (0.64, 1.05) 249.69 80 (31.15; 23, 138)
Alzheimer’s disease
 September . . . 0.84 0.69 (0.55, 0.82) 196.87 43 (19.62; 7, 79)
 October . . . 0.92 0.64 (0.50, 0.78) 189.71 79 (21.63; 39, 118)
Septicemia
 September . . . 0.37 0.22 (0.13, 0.30) 62.30 42 (12.56; 19, 66)
 October . . . 0.34 0.20 (0.12, 0.29) 60.30 39 (12.97; 15, 63)

Note. ARIMA = auto-regressive integrated moving average; CI = confidence interval. Causes of death were determined through International Classification of Diseases (10th Revision [ICD-10]; Geneva, Switzerland: World Health Organization; 2016, https://icd.who.int/browse10/2016/en) codes for heart disease (codes I00–I09, I11, I13, I20–I51), diabetes (codes E10–E14), Alzheimer’s disease (code G30), and septicemia (codes A40–A41).

a

All rates are per 1000 residents.

b

No excess deaths were observed in the other age groups.

c

No information was available on what constituted “other causes,” including ICD-10 codes.

Excess deaths were higher among men than women (632 and 579 deaths, respectively) and occurred only among people aged 60 years or older (1038 deaths). Most excess deaths were the result of heart disease (253 deaths), “other” causes (204 deaths), diabetes (195 deaths), Alzheimer’s disease (122 deaths), and septicemia (81 deaths).

DISCUSSION

Using time-series modeling of vital statistics data, we estimated 1205 excess deaths in Puerto Rico from Hurricane Maria. In comparison with similar studies, our estimate was higher than the 1139 deaths estimated by Santos-Lozada and Howard5 and slightly lower than the 1271 deaths estimated by Santos-Burgoa et al.,7 although both of these estimates overlapped with our confidence interval. Our estimate might differ because of the different time periods assessed (Appendix C). The strength of our approach was the use of time-series modeling to account for secular trends, seasonal patterns, and changes in population denominators.

Similar to our study, Santos-Burgoa et al.7 used a time-series model. Both studies revealed higher excess mortality among older people (82.6% of excess deaths among people aged 65 years or older7 vs 86.1% among people aged 60 years or older in our study) and men (1.3 men:women7 vs 1.1 in our study). Other studies have shown higher mortality among women after natural disasters.12 The similarities in our estimates provide further evidence that our ARIMA time-series model is capable of accurately predicting excess deaths resulting from Hurricane Maria. However, we added previously unavailable information about the top causes—heart disease, diabetes, Alzheimer’s, and septicemia—which were likely attributable to lack of access to treatment.4

PUBLIC HEALTH IMPLICATIONS

To our knowledge, this study is the first time-series analysis to examine the top causes of excess deaths attributable to Hurricane Maria in Puerto Rico. Our results can help authorities assess the scale of the disaster and plan an adequate allocation of resources for the island’s recovery. Moreover, understanding the causes of excess morality can aid in public health preparedness for future disasters.

CONFLICTS OF INTEREST

We declare no conflicts of interest.

HUMAN PARTICIPANT PROTECTION

No protocol approval was needed for this study because only publicly available de-identified data were used.

REFERENCES

  • 1.Checchi F, Roberts L. Documenting mortality in crises: what keeps us from doing better? PLoS Med. 2008;5(7):e146. doi: 10.1371/journal.pmed.0050146. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Choudhary E, Zane DF, Beasley C et al. Evaluation of active mortality surveillance system data for monitoring hurricane-related deaths—Texas, 2008. Prehosp Disaster Med. 2012;27(4):392–397. doi: 10.1017/S1049023X12000957. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Central Intelligence Agency. The world factbook: Puerto Rico. Available at: https://www.cia.gov/library/publications/the-world-factbook/geos/rq.html. Accessed March 12, 2019.
  • 4.Kishore N, Marqués D, Mahmud A et al. Mortality in Puerto Rico after Hurricane Maria. N Engl J Med. 2018;379(2):162–170. doi: 10.1056/NEJMsa1803972. [DOI] [PubMed] [Google Scholar]
  • 5.Santos-Lozada AR, Howard JT. Use of death counts from vital statistics to calculate excess deaths in Puerto Rico following Hurricane Maria. JAMA. 2018;320(14):1491–1493. doi: 10.1001/jama.2018.10929. [DOI] [PubMed] [Google Scholar]
  • 6.Land KC, Cantor D. ARIMA models of seasonal variation in US birth and death rates. Demography. 1983;20(4):541–568. [Google Scholar]
  • 7.Santos-Burgoa C, Sandberg J, Suarez E et al. Differential and persistent risk of excess mortality from Hurricane Maria in Puerto Rico: a time-series analysis. Lancet Planet Health. 2018;2(11):e478–e488. doi: 10.1016/S2542-5196(18)30209-2. [DOI] [PubMed] [Google Scholar]
  • 8.World Bank. Life expectancy at birth, total for Puerto Rico. Available at: https://fred.stlouisfed.org/series/SPDYNLE00INPRI. Accessed March 12, 2019.
  • 9.Sanchez Hernandez E, Morales J. Vital statistics annual report (general mortality, infant mortality, maternal mortality, and fetal mortality: 2009–2014) [in Spanish]. Available at: http://www.estadisticas.gobierno.pr/iepr/LinkClick.aspx?fileticket=sBWL1ofjM0g%3d&tabid=186. Accessed March 12, 2019.
  • 10.US Census Bureau. American Community Survey 5-Year Estimates, 2008–2017. Available at: http://factfinder.census.gov. Accessed March 12, 2019.
  • 11.Montgomery DC, Jennings CL, Kulahci M. Introduction to Time Series Analysis and Forecasting. Hoboken, NJ: John Wiley and Sons; 2008. [Google Scholar]
  • 12.Neumayer E, Plümper T. The gendered nature of natural disasters: the impact of catastrophic events on the gender gap in life expectancy, 1981–2002. Ann Am Assoc Geogr. 2007;97(3):551–566. [Google Scholar]

Articles from American Journal of Public Health are provided here courtesy of American Public Health Association

RESOURCES