Traditional Versus Contemporary Models for Identifying Causes of Maternal Death: A Population‐Based Study

ABSTRACT

Objective

To examine differences in cause‐of‐maternal‐death assignment based on underlying versus multiple causes of death.

Design

Cross‐sectional, population‐based study.

Setting

United States, 1999–2002 and 2018–2022.

Population

1633 maternal deaths in 1999–2002 and 1929 maternal deaths in 2018–2022.

Methods

Causes of death were identified based on the underlying cause of death and also based on multiple causes of death. The frequency of six selected obstetrical causes of death was quantified and ranked.

Results

Pre‐eclampsia (0.87 per 100 000 live births, 95% confidence interval [CI] 0.74–1.02) was the most common cause of maternal death in 2018–2022, when cause‐of‐death assignment was based on the underlying cause of death. Amniotic fluid embolism and cardiomyopathy tied for second rank, haemorrhage placed fourth and puerperal sepsis and uterine rupture tied for the fifth rank. Rankings based on multiple causes of death showed a different pattern: haemorrhage (1.13 per 100 000 live births, 95% CI 0.98–1.29) was the most common cause, followed by pre‐eclampsia, cardiomyopathy, amniotic fluid embolism, puerperal sepsis and uterine rupture. There was no significant correlation between the cause‐of‐maternal death rankings based on the underlying and multiple causes of death (correlation coefficient 0.62, 95% CI‐0.39, 0.95; p value 0.19) in 2018–2022. Cause‐of‐death rankings were significantly correlated under the two methods of cause‐of‐death assignment in 1999–2002 (correlation coefficient 0.83, 95% CI 0.05, 0.98; p value 0.04).

Conclusion

Basing cause‐of‐death assignment and ranking of the causes of maternal death using a multiple causes‐of‐death approach may better inform clinical and public health priorities for reducing maternal mortality.

1. Introduction

Maternal mortality remains a global public health priority, with substantial disparities across different regions and socio‐economic groups. Despite efforts to reduce maternal mortality, the global maternal mortality ratio (MMR) remains unacceptably high, particularly in low‐ and middle‐income countries [1, 2, 3]. Even in high‐income countries, concerns regarding rates, time trends and causes of preventable maternal death are regularly highlighted in both scientific journals and the lay media [4, 5, 6, 7, 8, 9, 10].

Accurate classification of maternal deaths and detailed information on cause‐specific maternal mortality are crucial for developing effective clinical and public health interventions [4, 11, 12]. International comparisons of maternal mortality and benchmarking are other helpful methods for spurring improvements in health services and public policy. However, between‐country and regional differences in definitions, assignment of causes of death (as direct or indirect), categorisation of overdose, suicides and homicides among pregnant and postpartum women (as maternal or accidental deaths), and other details sometimes hinder meaningful quantification and comparisons [13, 14, 15, 16].

One such controversy involves the assignment of the underlying cause of maternal death. The underlying cause of death is defined as ‘(a) the disease or injury which initiated the train of morbid events leading directly to death or (b) the circumstances of the accident or violence which produced the fatal injury’ [17]. Between‐country differences in underlying cause‐of‐death assignment were highlighted some years ago by the disagreement over the case of a woman with pre‐eclampsia, who underwent a caesarean delivery, suffered a massive haemorrhage and surgical complications and who subsequently died as a consequence [16]. Members of the Maternal Mortality Committee from the United Kingdom (UK) favoured assigning postpartum haemorrhage as the underlying cause of death, whereas members of the Dutch Maternal Mortality Committee insisted that pre‐eclampsia was the more appropriate choice. The UK classification focused on the lesson for clinical practice (which related to management of haemorrhage), while the Dutch classification identified the underlying pathology (specifically the indication that led to the caesarean delivery) [16]. The ensuing impasse led to a plea for recommendations regarding standard criteria for determining the correct point in the chain of morbid events that best serves to identify the underlying cause of death [16].

The imperative to identify a singular underlying cause of death is based on public health concepts rooted in traditional ideas regarding cause and effect [16, 17]. We hypothesised that cause‐of‐death assignment based on contemporary, multifactorial (multiple causes of death) models of causation [18, 19, 20, 21, 22] would provide better insights into the causes of maternal death, and thereby facilitate clinical and public health prevention programmes. We, therefore, carried out a study to explore differences between the underlying and multiple cause‐of‐death approaches to maternal cause‐of‐death assignment.

2. Methods

2.1. Data Sources

The study was based on maternal deaths and live births in the United States in two periods, namely, 1999–2002 and 2018–2022. These time periods were selected to provide insight into potentially evolving death certification practices. The period 2003 to 2017 was not included in the study because maternal deaths identified solely because of a positive pregnancy checkbox during these years had the causes of death reassigned to a pregnancy chapter code [4, 23, 24, 25, 26, 27]. This resulted in a large fraction of maternal deaths receiving non‐specific causes of death [4, 23, 24, 25, 26, 27].

Data on maternal deaths were obtained from the ‘multiple causes of death’ files, and data on live births were obtained from the natality files of the National Center for Health Statistics, Centers for Disease Control and Prevention (https://www.cdc.gov/nchs/data_access/vitalstatsonline.htm). The multiple causes‐of‐death files contain information transcribed from death certificates completed by the physician certifying the death, and include a single underlying cause of death [28], and up to 20 multiple causes of death [29]. Cause‐of‐death information in these files is documented using International Classification of Diseases (ICD‐10) codes [17, 30].

2.2. Identification of Maternal Deaths

Maternal deaths [17, 30] occurring during pregnancy or within 42 days of termination of the pregnancy were identified based on multiple cause‐of‐death information. All decedents with a pregnancy‐related cause of death listed among the multiple causes of death were first identified (ICD‐10 codes A34, O00‐O99). Late maternal deaths (O96 and O97) and those with an ‘accidental’ underlying cause of death (i.e., transport accidents, intentional self‐harm or assault; ICD‐10 codes V01‐V99, W00‐Y36, Y85‐Y89) were excluded. Cases with a positive pregnancy checkbox and no mention of a pregnancy‐related cause among the (originally documented) multiple causes of death were not identified as maternal deaths. This was done to avoid potential misclassification of maternal and incidental deaths [4, 23, 24, 25, 26, 27]. Also, such deaths had been reassigned a pregnancy‐related underlying cause of death from the pregnancy chapter, but in most cases, the new underlying cause of death was non‐specific (e.g., Other maternal diseases classifiable elsewhere) [4, 23, 24, 25, 26, 27].

2.3. Cause‐Specific Maternal Mortality

Maternal deaths due to six selected obstetric causes of death were identified and the associated cause‐specific maternal mortality ratios were calculated for the 1999–2002 and 2018–2022 periods [27]. The numbers of live births in the two periods served as the denominators for the maternal mortality ratios. The selected obstetric causes of death, which represent common and unambiguous causes of maternal death [27], included (i) Pre‐eclampsia (i.e., chronic hypertension with superimposed pre‐eclampsia, pre‐eclampsia and eclampsia based on the ICD‐10 codes O11, O14 and O15); (ii) Amniotic fluid embolism (ICD‐10 code O88.1); (iii) Cardiomyopathy (ICD‐10 code O90.3); (iv) Haemorrhage (i.e., haemorrhage due to placenta previa, placental abruption and postpartum haemorrhage based on ICD‐10 codes O44, O45, O72); (v) Puerperal sepsis and other puerperal infections (ICD‐10 codes O85 and O86); and (vi) Uterine rupture (ICD‐10 codes O71.0 for rupture of the uterus before the onset of labour and O71.1 for rupture of the uterus during labour).

2.4. Causes Based on the Underlying Cause of Death

Under this approach, the cause of maternal death was assigned based on the underlying cause of death. Thus, a maternal death with haemorrhage and pre‐eclampsia listed among the multiple causes of death, and haemorrhage identified as the underlying cause of death, would have been counted as a case of maternal death due to haemorrhage. On the other hand, if this case with haemorrhage and pre‐eclampsia listed as the multiple causes of death had pre‐eclampsia listed as the underlying cause of death, it would have been counted as a case of maternal death due to pre‐eclampsia.

2.5. Causes Based on the Multiple Causes of Death

The cause of maternal death was also assigned based on the multiple causes of death. Thus, a maternal death with haemorrhage and pre‐eclampsia listed among the multiple causes of death would have been counted as a maternal death due to haemorrhage and also as a maternal death due to pre‐eclampsia (irrespective of the condition chosen as the underlying cause of death).

2.6. Statistical Analysis

Cause‐specific maternal mortality ratios were calculated along with 95% confidence intervals (CI). Cause‐specific mortality ratios in the two study periods and under the two schemes for cause‐of‐death assignment were used to rank causes of death by frequency. Differences in ranking between the two cause‐of‐death assignment schemes were compared using Spearman's correlation coefficient (ρ). A two‐sided p value of < 0.05 was used to guide inference. Statistical analysis was carried out using SAS 9.4 (SAS Institute, Cary, NC).

2.7. Ethics Statement and Patient Involvement

Ethics approval was not sought because the anonymised data used in the study are public available. Patients were not involved in the design and interpretation of this study.

3. Results

The study population included 1633 maternal deaths and 16 065 890 live births in 1999–2002 (yielding a maternal mortality ratio of 10.2 per 100 000 live births); and 1929 maternal deaths and 18 484 949 live births in 2018–2022 (yielding a maternal mortality ratio of 10.4 per 100 000 live births). There were substantial changes in maternal characteristics between 1999–2002 and 2018–2022. The proportion of live births to women < 25 years of age decreased, while the proportion of live births to women of advanced maternal age (≥ 35 years) increased from 13.5% to 19.3%. Race/ethnicity distributions showed a decrease in live births to non‐Hispanic White women, an increase in live births to Hispanic women and women of other races and an approximately similar proportion of live births to non‐Hispanic Black women. The proportion of live births to women who smoked decreased, as did the proportion of triplet live births, while twin live births increased slightly and live births to women with chronic hypertension and pre‐pregnancy diabetes increased substantially.

Table 1 shows selected cause‐specific maternal mortality ratios based on the underlying cause‐of‐death and multiple causes‐of‐death formulations. In 1999–2002, the most common cause of death based on the underlying cause‐of‐death formulation was pre‐eclampsia, with a frequency of 1.52 maternal deaths per 100 000 live births. Amniotic fluid embolism (0.94 per 100 000 live births) was ranked second in terms of frequency, followed by cardiomyopathy, haemorrhage, puerperal sepsis and uterine rupture. Pre‐eclampsia was also the most common cause of death under the multiple causes‐of‐death formulation (1.82 per 100 000 live births), followed by haemorrhage (1.18 per 100 000 live births), amniotic fluid embolism, cardiomyopathy, puerperal sepsis and uterine rupture.

TABLE 1.

Numbers and maternal mortality ratios (MMR) of cause‐specific maternal death based on underlying and multiple causes of death, United States, 1999–2002 and 2018–2022.

Period and cause of maternal death	Number of live births	Maternal deaths by underlying cause of death		Maternal deaths by multiple cause of death
		Number	MMR/100000 live births (95% CI)	Number	MMR/100000 live births (95% CI)
1999–2002
Pre‐eclampsia	16 065 890	244	1.52 (1.33–1.72)	292	1.82 (1.61–2.04)
Amniotic fluid embolism	16 065 890	151	0.94 (0.80–1.10)	187	1.16 (1.00–1.34)
Cardiomyopathy	16 065 890	101	0.63 (0.51–0.77)	128	0.80 (0.66–0.95)
Haemorrhage	16 065 890	76	0.47 (0.37–0.59)	189	1.18 (1.01–1.36)
Puerperal sepsis	16 065 890	22	0.14 (0.09–0.20)	46	0.29 (0.21–0.38)
Uterine rupture	16 065 890	13	0.08 (0.04–0.14)	25	0.16 (0.10–0.23)
Maternal deaths (total) a	16 065 890	—	—	1633	10.2 (9.67–10.66)
2018–2022
Pre‐eclampsia	18 484 949	161	0.87 (0.74–1.02)	205	1.11 (0.96–1.27)
Amniotic fluid embolism	18 484 949	132	0.71 (0.60–0.85)	197	1.07 (0.92–1.23)
Cardiomyopathy	18 484 949	132	0.71 (0.60–0.85)	202	1.09 (0.95–1.26)
Haemorrhage	18 484 949	111	0.60 (0.49–0.72)	209	1.13 (0.98–1.29)
Puerperal sepsis	18 484 949	18	0.10 (0.06–0.20)	40	0.22 (0.15–0.29)
Uterine rupture	18 484 949	18	0.10 (0.06–0.20)	29	0.16 (0.11–0.23)
Maternal deaths (total) a	18 484 949	—	—	1929	10.4 (9.98–10.9)

Note: 95% CI denotes 95% confidence interval. Pre‐eclampsia deaths were identified using International Classification of Disease codes O11, O14 and O15 for chronic hypertension with superimposed pre‐eclampsia, pre‐eclampsia and eclampsia, respectively; Amniotic fluid embolism O88.1; Cardiomyopathy O90.3; Haemorrhage using the codes O44, O45, O72 for placenta previa, placental abruption and postpartum haemorrhage, respectively; Puerperal sepsis using O85 and O86 for puerperal sepsis and other puerperal infections, respectively; and uterine rupture using codes O71.0 (rupture of the uterus before the onset of labour) and O71.1 (rupture of the uterus during labour).

^a Maternal deaths identified based on the presence of a pregnancy‐related cause among the multiple causes of death (see Methods).

Table 1 also shows cause‐specific maternal mortality ratios in 2018–2022 and rankings based on the underlying cause of death. These showed a pattern similar to cause‐specific mortality ranks based on the underlying cause of death in 1999–2002. Pre‐eclampsia (0.87 per 100 000 live births) was the most common cause of maternal death, followed by amniotic fluid embolism and cardiomyopathy tied for second rank (0.71 per 100 000 live births), haemorrhage in fourth place and puerperal sepsis and uterine rupture tied for the fifth rank. However, rankings based on the multiple causes of death showed a different pattern: haemorrhage (1.13 per 100 000 live births) was the most common cause of maternal death, followed by pre‐eclampsia (1.11 per 100 000 live births), cardiomyopathy, amniotic fluid embolism, puerperal sepsis and uterine rupture (Table 1).

The rankings of the causes of death based on the underlying cause‐of‐death formulation were correlated with those based on the multiple cause‐of‐death formulation in 1999–2002 (ρ = 0.83, 95% confidence interval [CI] 0.06, 0.98; p value 0.04; Table 2). There was no correlation between the rankings based on the underlying and multiple cause‐of‐death formulations in 2018–2022 (ρ = 0.62, 95% CI‐0.39, 0.95; p value 0.19; Table 2).

TABLE 2.

Cause‐specific maternal mortality ratios (MMR) and ranks for selected causes of maternal death based on underlying and multiple causes of death, United States, 1999–2002 and 2018–2022.

Period and cause of maternal death	Underlying cause of maternal death		Multiple cause of death		Spearman's correlation coefficient (95% CI)	p
	MMR/100000 live births	Rank (by frequency)	MMR/100000 live births	Rank (by frequency)
1999–2002
Pre‐eclampsia	1.52	1	1.82	1	0.83	0.04
Amniotic fluid embolism	0.94	2	1.16	3	(0.06, 0.98)
Cardiomyopathy	0.63	3	0.80	4
Haemorrhage	0.47	4	1.18	2
Puerperal sepsis	0.14	5	0.29	5
Uterine rupture	0.08	6	0.16	6
2018–2022
Pre‐eclampsia	0.87	1	1.11	2	0.62 (−0.39, 0.95)	0.19
Amniotic fluid embolism	0.71	2	1.07	4
Cardiomyopathy	0.71	2	1.09	3
Haemorrhage	0.60	4	1.13	1
Puerperal sepsis	0.10	5	0.16	6
Uterine rupture	0.10	5	0.22	5

Abbreviation: CI, denotes confidence interval.

4. Discussion

4.1. Main Findings

Our study contrasted two systems of cause‐of‐death assignment by ranking selected obstetric causes of maternal death based on the underlying and multiple causes‐of‐death approaches.

Cause‐of‐death ranks under the two systems were not always significantly correlated. The three causes of death with the highest frequency in the United States in 2018–2022 (among the selected obstetric causes of death) were pre‐eclampsia, amniotic fluid embolism and cardiomyopathy when cause‐of‐death assignment was based on the underlying cause of death, while haemorrhage, pre‐eclampsia and cardiomyopathy were the causes of death with the highest frequency when cause‐of‐death assignment was based on multiple causes of death. Ranks of the six selected causes of maternal death based on the underlying and multiple causes of death were not significantly correlated in 2018–2022 (ρ = 0.62, 95% CI‐0.39, 0.95). This difference in rankings between the underlying and multiple cause of death approaches in 2018–2022 implies that a ranking of clinical and public health priorities based on causes of maternal death may vary depending on the approach used for assigning cause of death.

4.2. Strengths and Limitations

Our study hypothesis regarding the greater potential utility of the multiple causes‐of‐death framework was based on contemporary models of causation, as applied to the occurrence of the event of interest, namely, maternal death. The hypothesis regarding the preferred approach to cause‐of‐death assignment was tested using population‐based data on maternal mortality from the United States.

The limitations of our study include a reliance on death certificate data for identifying maternal deaths. It is well recognised that maternal mortality is underestimated, and maternal and non‐maternal deaths can be misclassified when ascertainment is based solely on death certificate data [31, 32, 33, 34, 35] However, this limitation does not invalidate our thesis as cause‐of‐death ranks will likely differ between the underlying and multiple cause‐of‐death approaches to cause‐of‐death assignment irrespective of the rigour with which causes of death are determined. Another limitation arises because of the use of maternal mortality data from the Nation Center of Health Statistics of the United States. There are concerns related to maternal death misclassification in these files because of problems that are associated with the pregnancy checkbox on death certificates [4, 23, 24, 25, 26, 27]. On the other hand, this data set is publicly available, and the listing of the underlying and multiple causes of death provides sufficient detail for addressing the question posed in this study. Also, we avoided misclassification due to pregnancy checkbox issues by restricting the identification of maternal deaths to those with a pregnancy‐related cause listed among the multiple causes of death. This strategy ensured fewer false positive cases of maternal death, although maternal mortality rates would have been underestimated [4, 23, 24, 25, 26, 27].

Our study ranked selected causes of obstetric death, given our focus on the cause‐of‐death assignment controversy between the Dutch and U.K. Maternal Mortality Committees [16]. These causes of death differ from the most common causes of maternal death in the United States reported by the state‐based Maternal Mortality Review Committees (namely, mental health conditions, cardiovascular conditions, infection and haemorrhage) [36], although caveats regarding the utility of multiple vs. underlying cause‐of‐death approaches to cause‐of‐death assignment apply to such causes of death as well.

4.3. Interpretation

As expected, the frequency of each cause of death was higher when cause‐of‐death assignment was based on multiple causes of death as opposed to the underlying cause of death. This was because women who died following two or more complications (e.g., pre‐eclampsia and haemorrhage) would have counted towards 2 or more causes of death (i.e., both pre‐eclampsia and haemorrhage) under the multiple causes of death framework but only under a single cause of death under the underlying cause‐of‐death framework (e.g., pre‐eclampsia or haemorrhage but not both).

Identifying an underlying cause of death, as part of cause‐of‐death certification, was formalised in the early part of the twentieth century. The focus on identifying a singular underlying cause of death was motivated, at least partly, by problems with tabulating multiple causes of death in mortality statistics [18]. The need to identify an underlying cause of death was also supported by the Henle‐Koch postulates, and the early twentieth century paradigm that all (infectious) diseases had a single universal cause [19]. With the subsequent rise of non‐communicable diseases, modern epidemiologic and medical concepts transitioned from the ‘monocausal disease model’, towards favouring a disease model with multifactorial causation [19]. Contemporary certification of (maternal) death involves the documentation of the immediate cause of death, and a causal chain of events that leads back to the underlying cause of death. Nevertheless, despite the modern commitment to multifactorial models of disease [20, 21, 22], tabulation of the causes of death and quantification of cause‐specific maternal mortality continues to rely largely on the singular underlying cause of death, rather than a pluralistic viewpoint based on multiple causes of death [37, 38, 39].

Routine cause of death reporting in most countries is based on a single underlying cause of death [37, 38, 39], although research reports and special studies are increasingly based on multiple cause‐of‐death analyses [40, 41, 42]. A recent study on deaths in the United States, 2011–2019 [42], showed that the most common underlying causes of death were heart disease, malignant neoplasms and chronic lower respiratory disease, while analysis based on the multiple causes of death showed that only 53%, 91% and 50% of deaths from these causes had been identified under the underlying cause‐of‐death analysis [42]. The literature also shows that some causes of all‐cause mortality listed on death certificates tend to have a high ratio of mentions among the multiple causes of death relative to selection as the underlying cause (e.g., septicaemia), while others have a greater tendency to be chosen as the underlying cause (e.g., motor vehicle accident) [40]. It is possible that the relatively infrequent mention of uterine rupture among the maternal deaths in our study reflects a preference for listing the rupture‐associated haemorrhage as the underlying cause of death.

Simplicity is the strongest argument for basing cause‐of‐death assignment on the underlying cause of death, while the lack of a comprehensive, albeit more complex, aetiologic perspective is the most significant weakness of this method. These advantages and limitations of cause‐of‐death assignment based on the underlying and multiple causes of death were identified as early as 1902 in the first Manual of the International Classification of the Causes of Death, which stated that ‘All systems of mortality statistics are compiled on the principle that the number of cases of death and causes of death must be the same. So long as this is the case, it is practically impossible to give a full view of the causes of death’ [18].

The correlation coefficients assessing concordance in ranks between the underlying and multiple cause‐of‐death approaches to cause‐of‐death assignment in 1999–2002 and 2018–2022 had wide confidence intervals and were not significantly different from each other. However, at issue is the difference in ranks between the two frameworks for assigning cause of death (as evident in the 2018–2022 analysis) and not the difference between correlation coefficients. Attributing deaths to multiple different causes results in the number of causes of death exceeding the number of deaths. Thus, differences in cause‐of‐death rankings between the underlying and multiple causes‐of‐death approaches will increase as the number of (multiple) causes of death assigned to each death increases. The mean number of (multiple) causes of maternal death on death certificates in the United States increased from 2.9 in 1999–2002 to 3.3 in 2018–2021, while those with only a single cause of death decreased from 17.7% to 14.2% over the same period [4]. This could explain the correlation in cause‐of‐death rankings based on underlying and multiple causes of death in 1999–2002 and the lack of such concordance in 2018–2021 (Table 2).

Cause‐of‐death assignment based on multiple causes of death provides an etiologic perspective that is consistent with modern models of causation. Each of the multiple causes of death is considered to be a component of the sufficient cause, with the absence/elimination of a single component deemed capable of preventing death [16, 17, 18]. Alternatively, all multiple causes of death are considered part of a causal chain of events that sequentially links the immediate cause of death with the preceding causes of death and ultimately the underlying cause of death. The removal of a single link/cause in this chain of events leading to death is deemed capable of preventing death [16, 17, 18]. However, as mentioned, this scheme of cause‐of‐death assignment introduces complexity as the sum of cause‐specific mortality rates exceeds the overall mortality rate. Nevertheless, this pluralistic perspective makes for more effective identification of priorities for clinical and public health intervention. With regard to the UK–Dutch controversy regarding the underlying cause of death in the case with both pre‐eclampsia and haemorrhage [12], death could have been avoided either by prevention or better management of pre‐eclampsia or by prevention or better management of the major obstetrical haemorrhage.

The value of both underlying and multiple cause‐of‐death models depends on the accuracy of data on the causes of death. Death certificates will not accurately capture all the causes of death in the absence of clinical, social and other details and context. Whereas analysis based on multiple causes‐of‐death may hold greater utility for maternal death for maternal mortality prevention initiatives, it remains constrained by the same data quality issues that affect single cause‐of‐death models.

5. Conclusion

Differences in identifying the event in the causal chain that best serves as the underlying cause of death make international comparisons of cause‐specific maternal mortality less meaningful. The contemporary focus on a singular underlying cause of death, while compelling in its simplicity, is not consistent with modern models of multifactorial causation. Prevention efforts would be more effective if more than one cause is identified as responsible for maternal death, which often involves both serious pregnancy conditions (such as pre‐eclampsia) and associated complications (such as major obstetrical haemorrhage). Targeting each individual critical cause within a chain of causation can result in more effective prevention initiatives. Future research should focus on multiple cause‐of‐death models in order to examine if such a perspective leads to better agreement in international cause‐of‐death classification systems and improves the efficacy of clinical and public health initiatives for preventing maternal death and reducing maternal mortality.

Author Contributions

Y.S. and K.S.J. proposed the study and contributed to the design. Y.S. carried out the analysis with help from S.L. and K.S.J., Y.S. is the guarantor. Y.S. drafted the first version of the manuscript. All authors contributed to interpreting the results. All authors contributed to the content and critical revision of the manuscript and approved the final version for submission. The corresponding author (Y.S.) attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.

Disclosure

Declaration of transparency: The lead author (YS) affirms that this manuscript is an honest, accurate and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are openly available in Mortality Multiple Cause Files at https://www.cdc.gov/nchs/data_access/vitalstatsonline.htm.