Maternal and Child Health Journal

, Volume 14, Issue 6, pp 838–850

Perinatal Periods of Risk: Analytic Preparation and Phase 1 Analytic Methods for Investigating Feto-Infant Mortality

Authors

    • Division of Reproductive HealthCenters for Disease Control and Prevention
    • CityMatCH, Department of PediatricsUniversity of Nebraska Medical Center
    • Division of Family Health ServicesFlorida Department of Health
  • Magda G. Peck
    • CityMatCH, Department of PediatricsUniversity of Nebraska Medical Center
  • Carol S. Gilbert
    • CityMatCH, Department of PediatricsUniversity of Nebraska Medical Center
  • Vera R. Haynatzka
    • CityMatCH, Department of PediatricsUniversity of Nebraska Medical Center
  • Thomas BryantIII
    • Duval County Health Department, Institute for Public Health Informatics and Research
Article

DOI: 10.1007/s10995-010-0625-4

Cite this article as:
Sappenfield, W.M., Peck, M.G., Gilbert, C.S. et al. Matern Child Health J (2010) 14: 838. doi:10.1007/s10995-010-0625-4

Abstract

The Perinatal Periods of Risk (PPOR) methods provide the necessary framework and tools for large urban communities to investigate feto-infant mortality problems. Adapted from the Periods of Risk model developed by Dr. Brian McCarthy, the six-stage PPOR approach includes epidemiologic methods to be used in conjunction with community planning processes. Stage 2 of the PPOR approach has three major analytic parts: Analytic Preparation, which involves acquiring, preparing, and assessing vital records files; Phase 1 Analysis, which identifies local opportunity gaps; and Phase 2 Analyses, which investigate the opportunity gaps to determine likely causes of feto-infant mortality and to suggest appropriate actions. This article describes the first two analytic parts of PPOR, including methods, innovative aspects, rationale, limitations, and a community example. In Analytic Preparation, study files are acquired and prepared and data quality is assessed. In Phase 1 Analysis, feto-infant mortality is estimated for four distinct perinatal risk periods defined by both birthweight and age at death. These mutually exclusive risk periods are labeled Maternal Health and Prematurity, Maternal Care, Newborn Care, and Infant Health to suggest primary areas of prevention. Disparities within the study community are identified by comparing geographic areas, subpopulations, and time periods. Excess mortality numbers and rates are estimated by comparing the study population to an optimal reference population. This excess mortality is described as the opportunity gap because it indicates where communities have the potential to make improvement.

Keywords

Perinatal periods of risk (PPOR)Feto-infant mortalityHealth disparities methodology

Introduction

Because the U.S. infant mortality rate is no longer decreasing appreciably [1], racial disparities in infant mortality persist [2], and infant mortality is a complex and multifaceted issue [3, 4], communities across the nation have sought guidance in analyzing, interpreting, and effectively using community level data to address the issue of infant mortality. The Perinatal Periods of Risk (PPOR) approach provides urban communities with a framework and tools to use in investigating and preventing feto-infant mortality [510].

The PPOR analytic approach developed for use in the U.S. was adapted from established methods used in other countries [11], and is intended to be applied within the context of the larger six-stage PPOR approach described elsewhere in this journal issue [12]. The Stage 2 analytic process begins after a community has completed the Stage 1 process of initiating or augmenting a community planning process and assessing its community and analytic readiness.

The PPOR analytic approach can be divided into three major parts (Table 1). Analytic Preparation focuses on acquiring, preparing, and assessing vital records files. Phase 1 Analysis estimates the excess mortality, or opportunity gaps, based on an optimal or most appropriate reference group for four mutually exclusive perinatal risk periods that contribute to feto-infant mortality. The four periods are defined by two major determinants: birthweight and age at death. Phase 2 Analyses are a series of in-depth community investigations of risk and preventive factors that potentially contribute to the excess mortality or opportunity gaps identified in Phase 1 Analysis.
Table 1

Perinatal periods of risk—analytic preparation and analytic methods

Analytic preparation

 1. Acquire and prepare vital records files for the defined study population

 2. Assess data quality

 3. Restrict study population by birthweight and gestational age

 4. Assess study sample size and re-define study population if needed

Phase 1 analysis—feto-infant mortality map and gaps

 1. Calculate numbers and rates for the feto-infant mortality map

 2. Compare different time periods, subpopulations and geographic areas

 3. Select reference populations

 4. Calculate excess mortality and identify opportunity gaps

Phase 2 analyses—further epidemiologic investigations

 1. Identify pathways or mechanisms for excess mortality

 2. Estimate prevalence of risk and preventive factors by type of mechanism

 3. Estimate the impact of the risk and preventive factors

The purpose of this article is to describe the first two parts of the PPOR analytic approach and to provide a community example of the calculations. The next article in this issue describes Phase 2 Analyses [13]. Additional information on PPOR, as well as related tools and resources, are available for public use from the CityMatCH website [14].

Analytic Preparation

Acquiring, preparing, and assessing the required vital records computer files are essential first steps to conducting PPOR analyses. Specific recommendations to help assure data quality and validity and usefulness of the results are described below.

Acquire and Prepare Vital Records Files for the Defined Study Population

Phase 1 PPOR analysis uses three primary vital records data files: fetal death certificates, linked birth and infant death certificates, and live birth certificates. Fetal death, death, and live birth certificate computer files are routinely created from electronic or paper certificates in every state and are linked annually by each state for submission to the National Center for Health Statistics (NCHS) at the Centers for Disease Control and Prevention (CDC) for inclusion in the CDC/NCHS Perinatal Mortality Data File [1]. The linked birth and infant death certificate is created by matching infant death certificates to live birth certificates manually through scanning records or electronically using computer algorithms, and differs from the death certificate file that is routinely used to calculate and report infant mortality rates.

The linked file should include those deaths that could not be linked to birth certificates. This file can be arranged and stored in one of two formats: death cohort or birth cohort. The death cohort (period data) contains all linked birth and infant deaths by year of death regardless of birth year. The birth cohort contains these linked events by year of birth regardless of death year.

Traditional infant mortality rates are calculated by dividing the number of infant deaths in a year by the number of live births in the same year. True infant mortality risks are estimated by dividing the number of infant deaths from a birth cohort by the number of live births for the same birth year. If timelier estimates are required, infant mortality risks can be approximated by dividing the number of infant deaths from a death cohort by the number of live births for the same year. These approximate risks can be biased if large changes in infant mortality, population, health systems or community context occur over a short period of time. However, they are frequently used because of their similarity to traditional infant mortality rates and because they can obviate the need to wait an additional calendar year to have a complete birth cohort. These considerations apply to PPOR analysis and affect the data request.

The PPOR study population is generally defined by time period and geo-political boundaries based on mother’s place of residence at the time of birth, and should include all fetal and infant deaths among resident women regardless of place of occurrence. Rates estimated using place of birth or death occurrence are generally distorted by factors that caused births and deaths to occur at that location. For example, hospitals with highly specialized intensive perinatal care services have high feto-infant mortality rates because they serve as referral centers for high-risk pregnant women and infants. Deaths which occur outside the community and state should generally be included in the analysis because community factors, such as access to care, can contribute to migration or seeking outside services before or after birth. These deaths should be assessed to examine whether specific factors or events related to these pregnancies and deaths need to be included as part of the community’s prevention planning.

Access to the three required vital records data files may be limited by law, regulation, or agency policy. Barriers to data access may include confidentiality issues, human subjects review requirements, data quality, and political sensitivities. Fetal death files have been especially difficult for local agencies to obtain. Most state agencies will require the submission of a formal data request for review and approval by the agency’s vital records staff and/or an institutional review board. The analyst must ensure that the files contain all the data elements required for Phase 1 Analysis. The analyst may also want to ensure that potential data elements for Phase 2 Analyses are requested at the same time (see Table 2). A more extensive protocol and data request will often be required to obtain the data elements and files for Phase 2 Analyses. Many communities underestimate the time, effort, and resources required to gain access to vital record files.
Table 2

PPOR data elements

Live birth file

Fetal death file

Linked birth/infant death filea

Required phase 1 analytic elements

 Birthweight

Birthweight

Age at death

 Birth year

Birth year

(Not linked to birth certificates)

 Gestational age

Gestational age

 

 Maternal age

Maternal age

 

 Maternal education

Maternal education

 

 Maternal Hispanic ethnicity

Maternal Hispanic ethnicity

 

 Maternal race

Maternal race

 

 Place of residence

Place of residence

 

Additional potential phase 2 analytic elements

 Birth hospital

Birth hospital

Autopsy

 Cesarean delivery

Cesarean delivery

Death circumstances

 Diabetes

Diabetes

Death hospital

 Gestational diabetes

Gestational diabetes

External injury causes

 Gestational weight gain

Gestational weight gain

Underlying cause of death

 Hypertension

Hypertension

 

 Infant transfer

Infant transfer

 

 Maternal height

Maternal height

 

 Maternal transfer

Maternal transfer

 

 Pay source

Pay source

 

 Perinatal hospital level

Perinatal hospital level

 

 Preeclampsia/eclampsia

Preeclampsia/eclampsia

 

 Prenatal care

Prenatal care

 

 Prepregnancy weight

Prepregnancy weight

 

 Previous preterm birth

Previous preterm birth

 

 Sexually transmitted diseases

Sexually transmitted diseases

 

 Smoking

Smoking

 
 

Prepartum death

 

aSame data elements obtained from the birth certificate are obtained from the birth certificate portion of the linked file

The study population must have at least 60 feto-infant deaths (sample size is explained in the section on Assess Study Sample Size) and generally must be defined before requesting the data files. The data request should be for the largest geo-political boundary and number of years that may be required. Urban communities interested in estimating the feto-infant mortality rates for a state reference group or other comparison population group may want to request the entire state data file. A codebook, or list of available data elements with detailed descriptions, definitions, and coding, will be an important aid in making a data request and conducting subsequent analyses.

Once obtained, the vital record files must be arranged with data elements coded, calculated, and/or edited for consistency to allow calculation of feto-infant mortality rates by maternal and infant characteristics. For example, prenatal care indicators, such as the Kotelchuck index [15], should be calculated using the same algorithm.

If the plan is to directly estimate infant mortality risk, death period files need to be converted to birth cohort files. Depending on the analyst’s preferences, these three files may be combined into one cohort study file of birth and death information including all births, fetal deaths, and infant deaths. This combined file can simplify programming for Phase 1 Analysis and facilitate multivariable analyses during Phase 2 Analyses. If data from different revisions of birth, death, and fetal death certificates are to be combined or compared, variables with revised definitions will need to be recoded for consistency.

Assess Data Quality

The first analysis should be an assessment of data quality. Quality should not be assumed because vital record data quality varies substantially by locality and over time [14]. A common problem identified by PPOR is the need to improve the quality of vital records reporting. To assess data quality, several procedures are recommended. First, compare the total number of live births, fetal deaths, infant deaths, and linked infant deaths in the data files with published totals. Sometimes differences are found because records are added to the linked statistical files used for analysis before and after published totals, because changes are made to records related to adoption, or because unnecessary duplicate records are created in the linkage process. Differences between the data in the files and published data should be identified, resolved, and explained.

Next, determine the numbers and percentages of live births with missing birthweight, unlinked infant deaths, infant deaths with missing birthweight, and fetal deaths with missing birthweight or gestational age. In most cities, live birth records have substantially fewer missing values than infant and fetal death records. Because age at death and birthweight or gestational age are generally required to complete a PPOR feto-infant mortality map, mortality events which are missing this information (including all unlinked infant deaths) cannot be included in the analysis. This exclusion falsely lowers the community’s mortality numbers and rates by the amount of missing data, which in some communities exceeds 10% of all deaths. This difference is a substantial bias as it is more than the decline in the U.S. infant mortality rate over the last 10 years.

We recommend testing for implausible or “out-of-range” birthweight, gestational age and birthweight/gestational age combinations. For example, because it is unlikely that a 1,000 gram live birth would be born at 38 weeks gestation, one would assume that at least one of the two data elements is incorrect [16, 17]. Values thought to be implausible should be treated as unknown or missing values. This increases the number and percentage of unknown values.

When large numbers of deaths are missing necessary data elements, one way to decrease the bias is to impute unknown values based on closely correlated data elements with known values. For example, birthweight can be imputed from gestational age and vice versa. Although imputation is not completely accurate, it may be preferable to completely excluding all unknown events from the analysis. Imputation should be considered if more than 5% of the birthweight or gestational age values are missing from fetal or infant death records; imputation is strongly recommended if more than 10% of the birthweight or gestational age values are missing. Imputation algorithms were developed for PPOR based on national NCHS data for the years 1995–1997, using median birthweight for a given gestational age. Fetal deaths at 32 weeks gestation or more are imputed to have birthweights of 1,500 g or more. Infant deaths at 31 weeks gestation or more also are imputed to have birthweights of 1,500 g or more. Fetal and infant deaths at gestational ages less than 32 and 31 weeks, respectively, are imputed to weigh less than 1,500 g. Fetal deaths at less than 24 weeks gestation and infant deaths at less than 22 weeks gestation are imputed to have birthweights of less than 500 g and are excluded from PPOR analysis (see section on Restrict the Study Population by Birthweight and Gestational Age). Other imputation procedures are available for use [16, 18, 19]. Imputation procedures used should be specified clearly.

The percentage of missing values for other data elements also should be evaluated. Although these missing values will not eliminate the vital event from the overall PPOR analysis, missing maternal information can bias the internal reference group and other comparisons. The bias can be large if the values are missing differentially. For example, if a larger percentage of maternal race, education, or age is missing among deaths than among live births, the calculated feto-infant mortality rate for the PPOR reference group will be artificially low. Even if non-differential, a large percentage of missing values for a data element limits the ability to study the effect of that factor.

Restrict the Study Population by Birthweight and Gestational Age

PPOR study populations are restricted by birthweight and gestational age to assure accuracy and comparability of vital statistics reporting both within and across communities. Based on extensive analysis of 1995–1997 NCHS data, we recommend restricting live births and infant deaths to birthweights of 500 g or more and restricting fetal deaths to birthweights of 500 g or more and gestational ages of 24 weeks or more. These restrictions are crucial because reporting lower birthweight and gestational age categories varies by state, region, physician, and hospital and can distort PPOR analyses [2023]. The birthweight distribution below 500 g among U.S. cities with populations of 250,000 varies widely using national vital records files. Restricting the analysis to births of 500 g or more addresses most of this urban variation. Birthweight restrictions, however, are not sufficient for fetal death reporting. When comparing fetal mortality rates for the largest U.S. cities, cities in states that report all fetal deaths, regardless of gestational age and birthweight, have higher fetal mortality rates at earlier gestational ages than cities in states that do not report all fetal deaths. This discrepancy persists until approximately 24 weeks gestation. This recommended cutoff is more inclusive than the cutoff of 28 weeks proposed by McCarthy and colleagues in the original WHO Periods of Risk model [11].

Feto-infant deaths with birthweights and gestational ages below the cutoffs described above frequently represent more than 25% of a community’s overall feto-infant mortality rate. While excluding such a large number of deaths can appear problematic, these extremely premature mortality events would be added to the PPOR risk period that in general already experiences the highest mortality rates and excess mortality in almost all U.S. communities. Including these events would only further emphasize the priority of addressing maternal health and prematurity issues in most communities and potentially distort the study of risk factors that contribute to mortality among the extremely premature. Communities interested in studying these extremely premature events should analyze the events as a separate category rather than including them as a part of the standard four PPOR risk periods. This approach permits study of these events separately and preserves the standard risk period structure for comparison of findings across communities.

Assess Study Sample Size

The last preparation step is to determine whether the number of feto-infant deaths in the study population is sufficient to conduct a statistically reliable PPOR analysis. When the number of deaths in a community is small, mortality rates tend to fluctuate substantially over time and do not provide stable, reliable information for policymaking and program planning. NCHS practice is to report rates when there are 20 or more deaths. This is described as a “convenient, if somewhat arbitrary, benchmark, below which rates are considered to be too statistically unreliable for presentation” [24]. We believe this risk period limit is conservative and would severely restrict the number of communities able to use PPOR methods as well as the number of potentially applicable reference populations for comparison.

To achieve statistically reliable results in Phase 1 Analysis, we recommend that a community have a cumulative minimum of 60 feto-infant deaths that meet the PPOR birthweight and gestational age limits, and at least 5,000 births over the same time period. This recommendation is based on a PPOR analysis of US cities with at least 250,000 or more population which showed that most US cities with 60 feto-infant deaths would generally have no fewer than 10 feto-infant deaths in any one risk period. If a community wants to further explore whether there is excess mortality within certain subpopulations of its community, a minimum of 60 deaths is needed in every population subgroup of study interest (e.g., teens, Native Americans, high risk neighborhoods, specific service areas such as Healthy Start Project areas, and local reference populations). Although the minimum recommendation is for 60 feto-infant deaths, a larger number of deaths would improve statistical reliability of the PPOR rates and would allow for more detailed Phase 2 Analyses. If a community does not have the minimum number of deaths overall or within the subpopulation of interest, the community should consider fetal and infant mortality reviews or some other investigative approach.

Some communities may need to combine several years of vital records data and/or neighboring geographic areas to reach sufficient numbers of feto-infant deaths. Combining time periods assumes that the community has experienced limited migration and demographic shifts and has had minimal changes in parental and infant characteristics, perinatal health systems, or other important community factors. Changes over time in perinatal health practices, services and systems may negate the potential usefulness of findings for time periods longer than 5 years. Neighboring geographic areas with similar populations, systems, and community characteristics may be combined, assuming the geographic unit of analysis is useful for community planning purposes. Some geographic combinations can be too large and/or too diverse to make meaningful interpretations of findings or to engage and mobilize communities.

PPOR analyses are not restricted to singleton live births. Multiple births contribute to a community’s overall feto-infant mortality rate and may contribute to an increasing rate over time or a population disparity. These births should be included in Phase 1 Analysis and further studied as part of Phase 2 Analyses.

Preparation of the data for PPOR analysis is very important and requires significant analytic resources. Access to the right data files and data elements is an essential, time-consuming process that often requires formal application and approvals. Assuring data quality, correct processing, and a sufficient number of deaths in the study population and subpopulations are essential for meaningful and reliable community results that can serve as the basis for action.

Phase 1 Analysis: Feto-Infant Mortality Map and Gaps

The PPOR Phase 1 Analysis estimates the community’s overall feto-infant mortality and divides the mortality into four periods of risk by use of a feto-infant mortality map (see Fig. 1). Maps can be compared across time, subpopulations and geographic areas to assess differences. Potential opportunity gaps are identified by comparison of study populations with an optimal reference group(s).
https://static-content.springer.com/image/art%3A10.1007%2Fs10995-010-0625-4/MediaObjects/10995_2010_625_Fig1_HTML.gif
Fig. 1

PPOR map of feto-infant mortality

Calculate Numbers and Rates for the Feto-Infant Mortality Map

With the study population defined and deaths restricted by birthweight and gestational age, the numbers and rates for the feto-infant mortality maps can be estimated. The PPOR map is constructed based on two major determinants that describe feto-infant mortality: age at death and birthweight. Age at death is partitioned into three discrete periods—fetal death (≥24 weeks gestation), neonatal death (birth to less than 28 days) and postneonatal death (28–364 days). Birthweight is partitioned into two discrete groups—500–1,499 and 1,500 g or more. Because the causes of feto-infant death were found to be very similar for the three age-at-death periods in the 500–1,499 birthweight group, the age-at-death periods for that birthweight range are combined into a single period of risk (see Fig. 1). Each feto-infant death in the study population is placed on the map into one of four perinatal periods of risk thus defined.

The four periods are color-coded for ease of communication, and labeled by the suggested primary prevention strategies that would most likely contribute to the health of the fetus or infant at this stage of development:
  1. (a)

    Maternal Health and Prematurity (blue)—fetal deaths of 500–1,499 g and ≥24 weeks and infant deaths of 500–1,499 g,

     
  2. (b)

    Maternal Care (pink)—fetal deaths of 1,500 g or more,

     
  3. (c)

    Newborn Care (yellow)—infant deaths of 1,500 g or more and less than 28 days of age, and

     
  4. (d)

    Infant Health (green)—infant deaths of 1,500 g or more and age between 28 days and less than 365 days.

     

Strategies to decrease mortality in the Maternal Health/Prematurity period might include improving preconception and prenatal health of the mother and prevention and treatment of prematurity. Strategies in the Maternal Care and Newborn Care periods might include improving routine preventive care, screening/assessment and referral, and high risk care and management, with the former focused on the mother and the latter focused on the infant. Infant Health strategies might include preventing sudden unexpected infant deaths and injuries, and improving access to care for life-threatening illnesses and conditions.

Different from the traditional birthweight-specific mortality rate, the mortality rate for each risk period is calculated for the study population by dividing the number of feto-infant deaths in that period by the total number of live births and fetal deaths. The total or overall rate for the study population equals the number of all infant and fetal deaths divided by the same total number of live births and fetal deaths. The number and rates for the four periods described above sum to the overall number and rate:
$$ {\text{Overall}}\,{\text{rate}} = {\frac{{\begin{array}{*{20}c} {{\text{Number}}\,{\text{of}}\,{\text{deaths}}} \\ {{\text{in}}\,{\text{period}}\,({\text{a}})} \\ \end{array} }}{{\begin{array}{*{20}c} {{\text{Total}}\,{\text{number}}\,{\text{of }}} \\ {{\text{live}}\,{\text{births}}\,{\text{and}}} \\ {{\text{fetal}}\,{\text{deaths}}} \\ \end{array} }}} \,+\, {\frac{{\begin{array}{*{20}c} {{\text{Number}}\,{\text{of}}\,{\text{deaths}}} \\ {{\text{in}}\,{\text{period}}\,({\text{b}})} \\ \end{array} }}{{\begin{array}{*{20}c} {{\text{Total}}\,{\text{number}}\,{\text{of }}} \\ {{\text{live}}\,{\text{births}}\,{\text{and}}} \\ {{\text{fetal}}\,{\text{deaths}}} \\ \end{array} }}} \,+\, {\frac{{\begin{array}{*{20}c} {{\text{Number}}\,{\text{of}}\,{\text{deaths}}} \\ {{\text{in}}\,{\text{period}}\,({\text{c}})} \\ \end{array} }}{{\begin{array}{*{20}c} {{\text{Total}}\,{\text{number}}\,{\text{of }}} \\ {{\text{live}}\,{\text{births}}\,{\text{and}}} \\ {{\text{fetal}}\,{\text{deaths}}} \\ \end{array} }}} \,+\, {\frac{{\begin{array}{*{20}c} {{\text{Number}}\,{\text{of}}\,{\text{deaths}}} \\ {{\text{in}}\,{\text{period}}\,({\text{d}})} \\ \end{array} }}{{\begin{array}{*{20}c} {{\text{Total}}\,{\text{number}}\,{\text{of }}} \\ {{\text{live}}\,{\text{births}}\,{\text{and}}} \\ {{\text{fetal}}\,{\text{deaths}}} \\ \end{array} }}} $$
The community can readily ascertain the contribution of each risk period to the overall rate because the period rates sum to the overall rate.

Compare Different Time Periods, Subpopulations and Geographic Areas

Feto-infant mortality maps can be calculated for different time periods, subpopulations and geographic areas, allowing comparisons that highlight disparities in overall rates and within the risk periods. Each map has its own set of numerators and its own denominator and will require a minimum of 60 feto-infant deaths. Estimating the rates for different time periods allows the community to assess whether rates are remaining constant, decreasing or increasing over time. Comparing the rates across different subpopulations reveals disparities, such as a black and white disparity in feto-infant mortality. Comparing the rates within and across geographic areas reveals geographic disparities, such as mortality rate differences between poor and wealthy neighborhoods. Caution is necessary in interpreting rates based on a small number of feto-infant deaths, especially when fewer than 10 deaths occur for a risk period or there are fewer than 60 deaths overall. If desired, the differences can be tested for statistical significance using chi-square or binomial tests.

Select Reference Population(s)

PPOR goes beyond many traditional statistical approaches to rate comparisons in that it estimates the excess mortality. To accomplish this, PPOR compares the mortality rates of the study population with the rates of an optimal reference population. The underlying assumption is, “if one population can experience better feto-infant mortality rates, then other populations should be able to attain the same rates.” Reference populations can be internal—a subpopulation within the urban community—or external—some appropriate population from outside the community.

An internal reference population should represent at least 15% of the overall study population to ensure realistic community representation and to have sufficient numbers. Generally, an external reference population is a population from another community, the state or the nation. Any reference population should be simply and reliably defined. It should have at least 60 feto-infant deaths for the study period of interest (preferably more), should experience optimal feto-infant mortality rates, and should be an acceptable comparison population for the community. The required data files and elements for Phase 1 and Phase 2 Analyses must be available for the reference group to allow further comparison, and must undergo the same analytic preparation as the study population data.

Based on our experience, a good reference population is non-Hispanic white women 20 or more years of age with 13 or more years of education. This population meets the above requirements for many U.S. communities. Some communities will need to use an external reference group because they do not have an internal reference population with at least 60 feto-infant deaths. Some communities also prefer an additional age limit, such as less than 35 years of age or additional years of education (e.g., 16 or more years or bachelor’s degree or higher). Other communities find their Asian or Latino/Hispanic subpopulations have lower mortality rates than the white subpopulation. One should explore how these changes might affect the findings as many of these do not have a large effect. Some communities have suggested developing a reference group based on preventable risk factors related to smoking cigarettes, alcohol consumption, early prenatal care, obesity, and pregnancy weight gain. However, this information is not always recorded completely or accurately on vital records. Moreover, these risk factors are more appropriately investigated in PPOR Phase 2 Analyses to help explain observed differences in mortality rates and provide advice on the ways to reduce disparities.

Ultimately, all reference populations for the PPOR analysis should be chosen strategically by the community for use in the planning process. Repeating the analysis using different reference populations, including both internal and external, can produce substantially different results and lead to a better understanding of differences in excess mortality and potential opportunity gaps. Information presented to the larger community, however, needs to be simplified, as analyses based on multiple references groups often provide too many numerical findings and can become confusing. Keeping the reference group simple and acceptable to the community is essential in using PPOR findings to change local practice.

Calculate Excess Mortality and Identify Opportunity Gaps

Once the reference populations have been selected, excess mortality rates and the excess number of deaths can be calculated. The overall excess mortality rate is simply the study population’s overall mortality rate minus the reference population’s overall mortality rate. Similarly, the period-specific excess mortality rates are the study population’s mortality rate for each specific risk period minus the reference population’s mortality rate for the same risk period.
$$ {\text{Population}}_{i} \,{\text{excess}}\,{\text{mortality}}\,{\text{rate}}_{j} = {\text{Population}}_{i} \,{\text{mortality}}\,{\text{rate}}_{j} -{\text{Reference}}\,{\text{population}}\,{\text{mortality}}\,{\text{rate}}_{j} $$
where i = overall or subpopulation and j = overall or perinatal risk period.
The number of excess deaths is frequently more understandable to community groups than the excess mortality rate. The number is estimated by multiplying the excess mortality rate of the study population by the number of births and fetal deaths for the same population. The number of excess deaths can be calculated overall and for each risk period.
$$ \begin{array}{*{20}c} {{\text{Population}}_{i} \,{\text{excess}}} \\ {{\text{number}}\,{\text{of}}\,{\text{deaths}}_{j} } \\ \end{array} = \begin{array}{*{20}c} {{\text{Population}}_{i} \,{\text{excess}}} \\ {{\text{mortality}}\,{\text{rate}}_{j} } \ \end{array} \,\times\, \begin{array}{*{20}c} {{\text{Population}}_{i} \,{\text{number}}\,{\text{of}}} \\ {{\text{live}}\,{\text{births}}\,{\text{and}}\,{\text{fetal}}\,{\text{deaths}}} \\ \end{array} $$
where i = overall or subpopulation and j = overall or perinatal period.

For example, the overall excess number of deaths among black infants is the excess mortality rate among black infants multiplied by the number of black live births and fetal deaths in the study population.

In the PPOR approach, excess mortality is used to identify opportunity gaps. Excess overall mortality can help a community identify subpopulations that contribute most to the excess, while excess period-specific mortality can help identify which risk periods are contributing most to a subpopulation’s excess deaths. Since each risk period is associated with distinct prevention strategies, the PPOR approach can guide the community toward interventions that will have the greatest local impact. Excess mortality varies across communities, by subpopulation and by risk period. Subpopulations and risk periods with the largest excess mortality or opportunity gaps become the foci for Phase 2 Analyses.

Phase 1 Analysis describes a community’s feto-infant mortality rate in such a way as to identify gaps that are opportunities for prevention. What is the potential for reducing the community’s feto-infant mortality rate? Which subpopulations have the most excess deaths? How do disparities contribute to the community’s excess mortality? Which risk periods are contributing most to the community’s feto-infant mortality excess? How is feto-infant mortality changing over time? What should be investigated further? The answers to these questions will guide and direct Phase 2 Analyses, which are discussed in the next article in this journal.

PPOR analyses have been conducted in more than 100 U.S. communities. The following example is presented to help guide users through Phase 1 calculations and to illustrate how communities can use the data to make lasting changes in programs, policies and institutions.

Example from Jacksonville, Florida

To illustrate Analytic Preparation and Phase 1 Analysis, we present an example from Jacksonville, Florida. The City of Jacksonville and Duval County define an urban area located in northeast Florida with a combined city/county government and jurisdiction. In 2003, 826,279 people resided in the 774 square mile city. The community is 29 percent African-American with a median age of 34.1 years. Hispanics make up less than five percent of the population. Despite the relative health of Jacksonville’s overall economy, significant pockets of poverty exist in the city. In 2002, 11 percent of the area’s residents had incomes below the federal poverty level, including 16.3 percent of families with children under age 18.

Jacksonville is served by the Northeast Florida Healthy Start Coalition, one of 32 local coalitions that comprise Florida’s Healthy Start Initiative. The Initiative, created by state law in 1991, ensures local leadership and planning for a system of care to promote healthy outcomes for pregnant women and infants [25, 26]. The system includes preconception counseling, prenatal care, postpartum care, delivery, infant care and targeted support services that address identified risks. As part of the state contract to receive public funding, local coalitions are required to develop service delivery plans based on a comprehensive community needs assessment. The Northeast Florida Healthy Start Coalition includes the PPOR approach as one of the core components of their required needs assessment [27].

The first step of Analytic Preparation in Jacksonville—acquiring and preparing data files for the study population—was far easier than in many U.S. urban communities. Once agency requirements are met, the Florida Department of Health routinely makes available to county health departments and local Healthy Start Coalitions the three required vital records files for PPOR analyses in a single combined birth cohort file. This file provides the core set of PPOR data elements for Phase 1 Analysis and most of the data elements for Phase 2 Analyses at a local level. The study population for this example was defined as all live births and fetal deaths to mothers who resided in Jacksonville at the time of birth between 1997 and 1999, inclusive.

The second preparation step—assessing data quality, including the number of unlinked infant deaths and deaths with missing data elements—determined that the number and percentage of unlinked infant deaths in Jacksonville was high, 45 of 358 infant deaths or 12.6% (See Table 3). The number and percentage of fetal deaths with unknown birthweight or gestational age was also high, 29 of 295 deaths or 9.8%. Both of these problems potentially bias the city’s feto-infant mortality estimates downward because these cases are excluded from the analysis. The estimated overall feto-infant mortality rate for Jacksonville was potentially 11.5% lower than the true rate because of excluding these events assuming the unknown events would not have been disproportionately excluded because of birthweight and gestational age restrictions. Although Jacksonville could have reduced this bias by imputing unknown values of birthweight for fetal deaths based on known gestational age, they chose not to do so because fewer than 10 percent of fetal deaths had to be excluded due to missing values.
Table 3

Number and percentage of unknowns for live births, fetal deaths, and infant deaths ineligible for PPOR study by data elements for Jacksonville/Duval County, Florida, 1997–1999

Infant and maternal characteristics

Live births

Infant deaths

Fetal deaths

Number

%

Number

%

Number

%

Total deaths

  

n = 358

   

 Unlinked deaths

NAa

NAa

45

12.6%

NAa

NAa

All births, linked infant deaths and fetal deaths

n = 36,161

 

n = 313

 

n = 295

 

 Birthweight

6

0.0

0

0.0

25

8.5

 Gestational age

21

0.1

0

0.0

9

3.1

 Gestational age or birthweight

24

0.1

0

0.0

29

9.8

 Age at death

NAa

NAa

0

0.0

NAa

NAa

All PPOR eligiblesb

n = 36,079

 

n = 240

 

n = 145

 

 Age

12

0.0

0

0.0

5

3.4

 Education

17

0.0

25

10.4

7

4.8

 Hispanic origin

26

0.1

0

0.0

0

0.0

 Race

501

1.4

0

0.0

0

0.0

 Any of the above

522

1.4

25

10.4

8

5.5

aNot applicable characteristic for live births and fetal deaths

bThese events meet PPOR study requirement and are not missing values for essential data elements

The percentages of infant and fetal deaths with unknown maternal education were also high. These unknowns falsely lowered the feto-infant mortality rates calculated by level of education because higher percentages are missing maternal education among death events than among live births. This affects the ability of the community to accurately estimate the number and rate for the reference population as well as to study the effect of maternal education on feto-infant mortality. Because methods of correcting for this bias are complex and require making questionable assumptions about the missing cases, Jacksonville chose not to impute.

The third step in Analytic Preparation was to restrict the study population by birthweight and gestational age. In Jacksonville, this resulted in the exclusion of an additional 58 live births, 73 infant deaths, and 121 fetal deaths. Ultimately, a total of 36,224 or 99.4% of live birth and fetal death events were eligible for use in the denominator in the PPOR analysis, and for the numerator, 240 or 67% of infant deaths and 145 or 49% of fetal deaths were eligible. More cases were excluded because of study restrictions than because of missing information.

The fourth step in Analytic Preparation—assessing the study sample size and study population—determined that Jacksonville, during 1997–1999, had a sufficient number of eligible feto-infant deaths to conduct PPOR analyses. Jacksonville had 385 eligible feto-infant deaths and had more than the minimum 60 deaths required to conduct PPOR analysis in two racial/ethnic subpopulations: non-Hispanic white (182 feto-infant deaths) and non-Hispanic black (190 feto-infant deaths).

In the first step of Phase 1 Analysis, the feto-infant mortality numbers, overall feto-infant mortality rate, and rates for the four risk periods were estimated. Of the 385 feto-infant deaths in Jacksonville, 180 occurred in the Maternal Health/Prematurity risk period, 81 in the Maternal Care period, 64 in the Infant Health period and 60 in the Newborn Care period. The overall feto-infant mortality rate was 10.6 deaths per 1,000 births (i.e., 385 feto-infant deaths divided by 36,224 fetal deaths and live births). The corresponding risk-period-specific mortality rates were 5.0 deaths per 1,000 births for Maternal Health / Prematurity, 2.2 for Maternal Care, 1.8 for Infant Health, and 1.7 for Newborn Care (See Table 4). Maternal Health/Prematurity was the leading risk period, accounting for 47% (180/385) of all feto-infant deaths in Jacksonville.
Table 4

Number and rates of feto-infant deaths and excess feto-infant deaths 1 by period of risk and population characteristics, Jacksonville/Duval County, Florida, 1997–1999

 

Feto-infant mortality rates

Numbers

Maternal health/prematurity

Maternal care

Newborn care

Infant health

Totald

Feto-infant deaths

Fetal deaths and live births

All

5.0

2.2

1.7

1.8

10.6

385

36,224

Race

 Non-Hispanic White

3.2

2.2

1.7

1.4

8.5

182

21,393

 Non-Hispanic Black

8.6

2.6

1.7

2.6

15.6

190

12,193

Time period

 All, 1997–1999

5.0

2.2

1.7

1.8

10.6

385

36,224

 All, 2000–2002

3.43

2.0

1.1

2.7

9.2

364

39,679

Reference rates

 Jacksonville reference

2.8

1.8

1.2

0.7

6.5

68

10,518

 Florida reference, 1998–2000 and

2.4

1.6

1.0

1.0

6.0

996

 

 US reference, 1998–2000a

2.2

1.5

1.1

1.0

5.9

23,054

 

Excess deaths using Jacksonville internal referenceb

 Allc

2.2

0.4

0.5

1.1

4.1

149

36,224

 Non-Hispanic White

0.4

0.4

0.5

0.7

2.0

43

21,393

 Non-Hispanic Black

5.8

0.8

0.5

1.9

9.1

111

12,193

aNational PPOR Data Tables, Table 6, www.citymatch.org

bFor each period and total, excess rates are the rates for the study population group minus the rates for the U.S. reference group. Number of excess deaths is the excess rate multiplied by the number of fetal deaths and live births

cThe number of excess death among all is less than the number for white and black because other has a negative number of excess deaths

dThe sum of the four periods may not exactly equal the total because of differences due to rounding

In the second step of Phase 1 Analysis, the PPOR rates were calculated for subpopulations (See Table 4), clearly demonstrating the black/white disparity in feto-infant mortality in Jacksonville. The rate for non-Hispanic blacks (15.6) was almost twice the rate for non-Hispanic whites (8.5). The disparity, the difference between these two values, was 7.1 deaths per 1,000. About three-fourths of the observed disparity occurred in the Maternal Health/Prematurity risk period: 5.4 deaths per 1,000 (8.6 for non-Hispanic blacks minus 3.2 for non-Hispanic whites). This represents 76% of the overall black/white disparity. Similarly, it was estimated that 17% of the deaths occurred in the Infant Health period and that the other two periods contributed very little to the disparity. The rates for other racial and ethnic populations were not reported for Jacksonville because these populations had fewer than 60 overall feto-infant deaths.

Temporal changes were assessed by comparing the PPOR map for 1997–1999 with the PPOR map for 2000–2002 (Table 4.). Over this time period, the overall feto-infant mortality rate in Jacksonville decreased by 1.46 feto-infant deaths per 1,000, from 10.6 to 9.2. The rate for the Maternal Health/Prematurity risk period decreased by 1.54 feto-infant deaths per 1,000, from 5.0 to 3.43. In contrast, the rate for Infant Health risk period increased to 0.93 deaths per 1,000 births, from 1.8 to 2.7 (P = .007). This was the only risk period where the feto-infant mortality rate actually increased.

For the third step in the Phase 1 Analysis, Jacksonville was able to use several different reference populations to calculate excess mortality (Table 4). A national reference group—an external reference population of white non-Hispanic women with 20 or more years of age and 13 or more years of education who were US residents—was provided by CityMatCH [28]. A second external reference group was created by applying the same criteria to Florida residents. Jacksonville’s local, internal reference population of white women in Duval County with 20 or more years of age and 13 or more years of education had higher mortality rates than either the Florida or U.S. reference group in every risk period except Infant Health. The internal reference group provided more easily attainable goals, which the community preferred in this case.

For any study population or reference group, the excess mortality rate for the fourth step in Phase 1 Analysis is calculated by subtraction. For example, the overall rate for Jacksonville of 10.6 minus the Jacksonville internal reference rate of 6.5 equals the overall excess mortality rate of 4.1 feto-infant deaths per 1,000 births (Table 4). The overall excess number of feto-infant deaths was calculated by multiplying the excess mortality rate of 4.1 per 1,000 by the number of live births and fetal deaths in Jacksonville, 36,224, which equals 149 excess deaths. Blacks experienced both a higher excess mortality rate and more excess deaths. The contribution of each risk period to excess mortality varied by race (Fig. 2). Maternal Health/Prematurity contributed 64% of black excess mortality and only 20% of white excess mortality. Infant Health contributed 22% for blacks and 39% for whites. Infant Health was the second highest contributor for blacks and the highest contributor for whites.
https://static-content.springer.com/image/art%3A10.1007%2Fs10995-010-0625-4/MediaObjects/10995_2010_625_Fig2_HTML.gif
Fig. 2

Excess feto-infant mortality using internal reference group, overall and by race, Jacksonville/Duval, Florida, 1997–1999

In summary, Jacksonville identified two high risk periods in their community using PPOR Phase 1 Analysis: Maternal Health/Prematurity, and Infant Health. The black/white mortality disparity was the biggest opportunity gap for improving black and overall feto-infant mortality in Jacksonville. These opportunity gaps became the focus of Jacksonville’s PPOR Phase 2 Analyses. Data quality issues were also identified in Jacksonville with a high frequency of unlinked infant deaths, missing birthweight on fetal death certificates, and missing maternal education on infant and fetal death certificates. Improving data quality was added as one of the community planning priorities.

Discussion

Perinatal Periods of Risk analytic methods have been used by many communities across the United States. Although more sophisticated epidemiological approaches exist, the alternatives often are not well suited for examining fetal and infant mortality rates at the community level. PPOR offers several clear strengths. It offers a simple systematic analytic approach that can be applied by communities with limited analytic skills and resources. The analysis is designed to extract as much useful information as possible from a small number of community mortality events using existing vital records and other data sources. PPOR analysis includes fetal deaths, which in many communities outnumber infant deaths. The approach also provides a visual summary that facilitates communicating the feto-infant mortality problem to the community at large. PPOR Phase 1 identifies which of four perinatal periods of risk contributes most to a community’s feto-infant mortality. It defines the opportunity gap by estimating the potential for reduction using reference groups, allowing a community to prioritize risk periods and subpopulations for further investigation. Phase 1 Analysis stops short of identifying factors that contribute to these gaps and therefore does not provide sufficient information for selection of appropriate interventions. PPOR Phase 2 Analyses can be used along with other information to accomplish this critical step.

A major strength of PPOR is the ability to readily describe the black-white gap and other disparities in communities in terms of both excess mortality rates and numbers of potentially preventable deaths [8]. In most localities, two common patterns appear: (1) excess mortality is largest in the Maternal Health/Prematurity period of risk, and (2) black excess mortality is much larger than white excess mortality. Much of this latter excess in racial disparity occurs in the Maternal Health/Prematurity period. These findings are consistent with analyses of fetal and infant mortality patterns that use other methods [29]. Some communities also find substantial excess mortality in the Infant Health period. Many of these deaths are preventable with currently known effective strategies.

One weakness of the PPOR approach is that many communities fail to continue their PPOR investigation after Phase 1. This can happen because of inadequate resources to conduct further analyses, changes in community or agency priorities, staffing changes, limited access to the necessary data files, or a community’s desire to move directly to action based on Phase 1 results. Identifying the risk period that provides the largest opportunity gap in Phase 1 is necessary but does not provide sufficient information for community planning. Further information about the actual causes, preventive/risk factors, and the community are needed to formulate an effective prevention plan and intervention measures. Although Phase 1 analytic results may be useful for surveillance, we do not recommend that communities undertake the PPOR community-based process unless they plan to complete both Phase 1 and Phase 2 analyses.

PPOR Analysis encounters the same statistical limitations as other investigative approaches. The small number of feto-infant deaths in many communities, both overall and among subpopulations, limits the information that can be gained through epidemiological studies. The more detailed the investigation needs to be, the more deaths that are required for precise answers. Many communities will have too few events to fully utilize the PPOR approach.

The quality and content of vital records is another limitation [2123, 3033]. Although PPOR provides quantifiable results quickly, the answers are only as good as the data sources. In some U.S. communities, the quality of vital records information is poor because of a large number of unreported fetal or infant deaths, unlinked infant deaths, or deaths with incomplete information or inaccurate reporting. In many communities, PPOR has increased the community’s political will to improve the quality of event collection and vital record reporting.

To be useful to communities with a small number of deaths, the PPOR approach simplifies perinatal issues and in some ways may oversimplify them. The complex interrelationships between low birthweight and prematurity cannot be fully addressed with a simple 1,500 g birthweight cutoff [29]. In reality, the four risk periods do not completely align with the labeled prevention strategies. Actual causes of death, especially those which occur near cutoff points, may not be typical for that risk period and thus may not be investigated. For example, the death of a normal birthweight baby that is actually due to perinatal conditions but occurs after the first 28 days of life will be classified in the Infant Health risk period, where risk factors for perinatal conditions are not usually considered. Communities could simply examine the prevalence of major preventable risk factors among all births and deaths. However, many risk factors contribute to mortality and disparity; community level resources and statistical power are often limited. Even if many factors are examined, it is very difficult for communities to put the findings into perspective, prioritize issues, and understand what interventions to choose. Connecting the risk and preventive factors to causation and mortality using PPOR helps communities allocate both analytic and programmatic resources.

This article has described the theoretical framework of the PPOR analytic approach and detailed the steps of data preparation and the first phase of analysis. It has illustrated the steps using an example from Jacksonville, Florida, one of many cities using PPOR. By following the steps covered in this article, a community can use its local vital records data to identify subpopulations and periods of risk with large excess mortality and to describe local health disparities. Identifying opportunity gaps narrows the realm of possibilities for preventive action by focusing on particular risk periods and study populations. But, many known possible mortality contributors and risk factors can contribute to these findings. Further analysis is necessary to determine which risk and preventive factors are likely to be contributing most to excess mortality in this community. Phase 2 Analyses [13], detailed in the following article, allow a community to investigate many of the reasons for the opportunity gaps and to choose the interventions most likely to have a positive local impact.

Acknowledgments

Thanks to Dr. Brian McCarthy for developing and sharing his original methods and encouraging us to modify the approach for use in U.S. cities. Thanks to the Perinatal Periods of Risk City Teams, Pat Simpson and Jennifer Skala for their help in developing and adapting the methods for use in urban communities. Special thanks to Dr. Milton Kotelchuck and Dr. Laurin Kasehagen for their advice in writing these articles for publication. This work was supported in part by the following Cooperative Agreements: Merging Research and Practice for Urban Child Health—TS-283-14/16 (under CDC Cooperative Agreement U50/CCU300860); Building Urban MCH Capacity—TS 0922 (under CDC Cooperative Agreement U50/CCU300860); Toward Greater Science Use in Urban Health Agencies—TS-1337 (under CDC Cooperative Agreement U50/CCU300860); and the Maternal, Infant, and Reproductive Health: Science-Based Capacity Building for Major Urban Public Health Agencies (5 U65 DP724969-05) between CityMatCH at the University of Nebraska Medical Center and the Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Division of Reproductive Health, with supplemental support from the National Center for Birth Defects and Developmental Disabilities, and the Health Resources and Services Administration, Maternal and Child Health Bureau. Additional support was provided by the National March of Dimes Birth Defects Foundation, and the University of Nebraska Medical Center, Department of Pediatrics.

Copyright information

© US Government 2010