Evaluating the impact of expanding the number of diagnosis codes reported in inpatient discharge databases on the counts and rates of birth defects

Abstract

Objective

Public health surveillance programs worldwide implement a variety of case-finding strategies, and many rely at least in part on International Classification of Diseases (ICD)-based diagnostic codes in administrative and clinical databases. Over time, state- and national-level hospital discharge databases have been expanding the number of reported diagnosis code fields. This study aimed to evaluate the impact of these expansions on frequencies and rates of major birth defects, and the classification of birth defects as isolated vs multiple.

Methods

We used state-level 2006-2013 Florida Birth Defects Registry data and 2009-2012 data from a nationally representative database (Kids’ Inpatient Database). We generated data under different scenarios by varying the number of diagnosis code fields available, and comparing counts and rates of major birth defects generated under each scenario.

Results

The expansion from 10 to 31 diagnosis code fields improved ascertainment by preventing the loss of 1 in every 40 birth defect cases with defect-related diagnoses appearing only in code positions 11 to 31. Although there was variation by birth defect, the largest impact of the expansion tended to occur for less severe birth defects diagnosed in sicker infants. When restricting to fewer codes, not only were fewer cases diagnosed, but more were classified as being isolated due to the inability to capture co-occurring defects.

Conclusion

Our findings encourage additional research for other health outcomes in patients of all ages. Other disease registries rely at least in part on diagnostic codes documented by healthcare providers in their case-finding activities, irrespective of ascertainment protocols, making routine investigation of these databases essential.

BACKGROUND AND SIGNIFICANCE

Public health surveillance programs and disease registries worldwide implement a variety of strategies to identify cases [persons with the condition(s) of interest]. These ascertainment strategies lie on a continuum from “passive” to “active.” Passive ascertainment relies on established case reports or identification of cases using secondary data sources, including linkage of administrative and clinical databases and the use of diagnostic coding systems. Active case ascertainment is a labor-intensive process in which staff find cases through direct review of primary data sources (eg, medical records, hospital/nursery logs); however, such methods result in higher completeness and accuracy relative to more passive approaches. The choice of ascertainment strategy is often determined by a confluence of factors, including surveillance goals, program funding, staff expertise and resources, and statutory authority for surveillance.^1–3

In the United States, over 60% of state-based birth defects surveillance programs submitting population-based prevalence data to the National Birth Defects Prevention Network (NBDPN) and the Centers for Disease Control and Prevention (CDC) report using a primarily passive case-finding methodology.⁴ The Florida Birth Defects Registry (FBDR) manages the second-largest population-based passive surveillance system in the United States, relying exclusively on the linkage of administrative databases and International Classification of Diseases, 9th edition, Clinical Modification (ICD-9-CM) and ICD-10 diagnostic codes to identify birth defects for ascertainment and reporting. Similar to other passive birth defects programs, the inpatient hospital discharge database is the primary data source for identifying cases, which captures uniquely or jointly over 93% of all cases with a major birth defect in the FBDR, and without which the FBDR would miss 75% of reported cases.⁵ As such, changes to the inpatient database may influence patterns of case reporting. In 2006, the inpatient hospital discharge database in Florida expanded the number of fields capable of capturing secondary diagnosis codes. Similar expansions in the number of diagnostic code fields also occurred in nationally representative inpatient databases, including the Healthcare Cost and Utilization Project’s Kids’ Inpatient Database (HCUP-KID).

Previous studies demonstrate that the quality of passive surveillance systems can vary greatly based on the types of administrative datasets included, the linking methodology employed, and the age by which diagnoses are made to be ascertained as a case.^6–8 However, despite its widespread occurrence in administrative databases, few studies have investigated the effect of increasing the number of diagnostic coding fields available in these databases on a registry’s completeness of ascertainment,^9–13 and to our knowledge, no articles have specifically researched birth defects registries. This study investigates the impact of the number of diagnosis codes available for use on the counts and rates of major birth defects in both the FBDR (state level) and HCUP-KID (national level), as well as the effect of the expansion on the classification of defects as isolated or multiple (co-occurring birth defects).

MATERIALS AND METHODS

Study design, data source, and sample

Data from the FBDR and HCUP were used for this study. On July 4, 1999, birth defects were added to the list of reportable conditions that may significantly impact health in Florida.¹⁴ The FBDR, Florida’s statewide, population-based surveillance program, was subsequently created by the Florida Department of Health (DOH) to protect, promote, and improve the health of people in Florida by detecting, investigating, and preventing birth defects. To be defined as a case in the FBDR, 3 inclusion criteria must be met: 1) biological mother is a Florida resident; 2) infant is born alive and is diagnosed in the first year of life with one or more structural, genetic, or other specified birth outcomes that can adversely affect health and development; and 3) infant was delivered on or after January 1, 1998.¹⁵ For reporting of birth defect counts and prevalence rates, the FBDR relies on a passive case ascertainment methodology that involves the acquisition, cleaning, and linking of various administrative data sources. First, the underlying source population from which cases arise is defined as Florida resident birth certificate records; in-state birth to Florida residents are added to out-of-state births to Florida residents, the latter made possible by interstate data sharing of birth certificate data. Then, using a hierarchical, deterministic data linkage strategy (as described previously),¹⁶ supplemental data sources with administrative and clinical information, including hospital discharge data, are linked to birth certificate records. Once linked to a birth certificate, each record from each of these supplemental sources contains ICD-9-CM diagnosis codes, or ICD-10 codes for infant death certificates, which are used to identify birth defects. For a given annual birth cohort, the FBDR constitutes an unduplicated inventory of infants that have 1 or more “included” birth defect codes (primarily ICD-9-CM code range 740-759 and ICD-10 “Q” codes). If an infant has multiple birth defects, all are documented.

The second source of data for this study is the HCUP-KID. HCUP is sponsored by the Agency for Healthcare Research and Quality and maintains nationally representative healthcare databases for research and policy analyses. HCUP has the largest multi-year, all-payer, encounter-level discharge data in the United States. The KID is a sample of pediatric inpatient records that allows researchers to study medical conditions and procedures specific to children. The KID is cross-sectional, generated every 3 years, and is weighted to allow generation of national counts and prevalence estimates.

At the state level, between 1998 and 2005, the FBDR’s primary database for ascertaining cases, the inpatient hospital discharge database, captured the principal ICD-9-CM diagnosis code and up to 9 additional (secondary) diagnosis codes. In 2006, the number of fields capable of capturing diagnosis codes was expanded from 10 (principal plus 9 secondary codes) to 31 (principal plus 30 secondary codes). Nationally, in 2009, a similar expansion occurred in the KID, which increased the number of diagnosis code fields captured and made available for analysis from 15 to 25 in 2009. Although each state agency serving as the repository for discharge data, as well as HCUP, implements quality control procedures¹⁷ to identify and exclude impossible ICD-9-CM codes (eg, pregnancy-related diagnosis for someone aged 4 years), we made no modifications to the ordering of diagnosis code fields as reported by each data source. The analytic timeframe for all analyses was restricted to the 2006-2013 FBDR and 2009 and 2012 KID, as these years occurred after the expansion of the number of diagnosis code fields in each data source and therefore allowed for comparisons under the different scenarios described below. All analyses preceded the implementation of ICD-10 in hospital discharge data in the United States.

Statistical analysis

All analyses were performed separately for the FBDR and the KID. Our study’s primary objective was to investigate the impact of the expansion of the number of available inpatient diagnosis code fields in state and national databases on the estimated count and live birth prevalence of birth defects. First, to better understand the relative position of individual ICD-9-CM codes included in the FBDR case definition, for each code, we calculated the percentage of all hospitalization records in which the code appeared in the first 10 code fields (vs code fields outside the first 10). Second, data from the FBDR and the KID were used to generate hypothetical annual surveillance data under 2 primary scenarios, and for each scenario, we calculated the number of cases and live birth prevalence rates (cases per 10 000 live births) for specific birth defects. As the actual change made to the inpatient database in the FBDR was a transition from 10 to 31 codes, we assessed whether the impact of the expansion resulted in statistically significant differences by determining whether prevalence estimates based on only the first 10 codes (as it was from 1998 to 2005) fell within or outside the 95% confidence interval of the “complete registry” based on 31 codes. For the KID, we compared estimates using the first 10 codes to estimates using all 25 codes. Each birth defect was defined by the presence of either one ICD-9-CM code (eg, Down syndrome: 758.0), or more often, any 1 of multiple ICD-9-CM codes indicative of the same type of defect (eg, limb reduction defects: 755.20-29, 755.30-39, 755.4). Next, we calculated the percentage of cases that would have been missed in each code-restricted scenario relative to the complete inpatient registry (using all available codes).

In addition to a scenario that reflected the actual number of diagnosis codes available in the FBDR prior to the code expansion (10), we also considered additional scenarios that were based on differing numbers of diagnosis code fields, starting with the full number of code fields currently available (FBDR: 31, KID: 25) and decreasing the number of codes used to 25 (FBDR only), 20, 15, and 5, and also using only the principal diagnosis from each hospitalization. For each scenario, we then scanned available diagnostic codes for birth defects, using the FBDR case definition, and compared counts and rates of major birth defects generated under each scenario to the “complete” surveillance data (using 31 fields for the FBDR, 25 fields for the KID).

Last, as failure to ascertain specific birth defects using ICD-9-CM codes not only affects that defect but also whether co-occurring defects are captured, we also assessed the impact of the number of available diagnosis code fields on the classification of defects as “isolated” or “multiple.” To convey this, the percentage of isolated cases of specific birth defects that was calculated using only the first 10 diagnostic code fields was compared to the percentage of cases that are classified as isolated using all available fields. To classify a birth defect as “multiple,” we used a computer algorithm to identify infants who were diagnosed with the listed birth defect and another birth defect from a different body system. Infants with multiple defects from the same body system, but not another body system, would be classified as “isolated.” The impact on isolated/multiple classification was restricted to the FBDR data, as the KID is cross-sectional and does not allow for the longitudinal followup that is important in capturing co-occurring defects that are not all present at birth.

For state-level analyses, to remove the potentially sizeable influence of other source datasets in the FBDR, counts and prevalence rates were calculated using only those ICD-9-CM diagnosis code fields in the inpatient database. All inferential tests were 2-tailed, with a 5% type I error rate, and analyses were conducted using SAS software (version 9.4; SAS Institute, Inc, Cary, NC). As this project constitutes an evaluation of the FBDR data whose reporting and surveillance are under the authority of Florida Statute 381.0031, this study did not require IRB determination.

RESULTS

Between 2006 and 2013, there were more than 18 000 live-born infants in the FBDR each year with 1 or more of the birth defects included in this study. Nationally, in 2009 and 2012, there were an estimated 114 559 annual hospitalizations to children under 1 year of age with a diagnosis of an included defect. Table 1 compares case counts and prevalence rates of selected birth defects when calculated using 10 vs the complete number of diagnosis code fields for the FBDR and KID. Overall, restriction to only the first 10 codes would have resulted in failure to identify 2.5% (786 of 31 485) of all cases in the FBDR, ranging from 0% missed for anencephalus to 22.2% (32 of 144 cases) for anophthalmia/microphthalmia. However, for 11 specific defects, the estimated prevalence using only the first 10 codes would have been statistically significantly different from the prevalence estimated using all 31 fields. For the KID, the percentage of cases that would have been missed range from 0.6% (4 of 630 cases) for anencephalus to 20.7% (234 of 1130 cases) for anophthalmia/microphthalmia. In addition, the prevalence for 24 birth defects when using the first 10 codes was statistically significantly different from the prevalence when all 25 code fields were used.

Table 1.

Comparison of case counts and prevalence rates of selected birth defects^a, calculated exclusively using 10 vs 31 diagnosis codes from inpatient hospitalization records^b, Florida Birth Defects Registry, 2006-2013, and using 10 vs 25 diagnosis codes from inpatient hospitalization records^b, KID Inpatient Database, Healthcare Cost and Utilization Project, 2009, 2012

Category/defect	FBDR				HCUP KID
	Cases reported		Reported rates and confidence intervals (per 10 000 Florida live births)		Cases reported		Reported rates and confidence intervals (per 10 000 hospitalizations)
	using all 31 codes	using first 10 codes	using all 31 codes	using first 10 codes	using all 25 codes	using first 10 codes	using all 25 codes	using first 10 codes
Central Nervous System
Anencephalus	83	83	0.47 (0.38, 0.58)	0.47 (0.38, 0.58)	630	626	0.67 (0.62, 0.72)	0.66 (0.61, 0.72)
Encephalocele	149	145	0.84 (0.72, 0.99)	0.82 (0.69, 0.96)	1005	984	1.06 (1.00, 1.13)	1.04 (0.98, 1.11)
Holoprosencephaly	740	703	4.17 (3.88, 4.48)	3.96 (3.68, 4.27)	7156	6770	7.56 (7.39, 7.74)	7.15 (6.98, 7.33)
Spina bifida without anencephalus	465	457	2.62 (2.39, 2.87)	2.58 (2.35, 2.82)	5795	5680	6.12 (5.97, 6.28)	6.00 (5.85, 6.16)
Eye
Aniridia	c	c	c	c	104	93	0.11 (0.09, 0.13)	0.10 (0.08, 0.12)
Anophthalmia/ microphthalmia	144	112	0.81 (0.69, 0.96)	0.63 (0.52, 0.76)	1130	896	1.19 (1.13, 1.27)	0.95 (0.89, 1.01)
Congenital cataract	156	125	0.88 (0.75, 1.03)	0.70 (0.59, 0.84)	1128	953	1.19 (1.12, 1.26)	1.01 (0.94, 1.07)
Ear
Anotia/microtia	125	98	0.70 (0.59, 0.84)	0.55 (0.45, 0.67)	1242	1047	1.31 (1.24, 1.39)	1.11 (1.04, 1.18)
Cardiovascular
Primary CCHD	2222	2205	12.53 (12.02, 13.06)	12.43 (11.93, 12.96)	25 054	24 586	26.47 (26.15, 26.80)	25.98 (25.65, 26.30)
Secondary CCHD	1834	1801	10.34 (9.88, 10.83)	10.16 (9.70, 10.64)	16 287	15 740	17.21 (16.95, 17.47)	16.63 (16.37, 16.89)
Primary or secondary CCHD	3477	3449	19.61 (18.97, 20.27)	19.45 (18.81, 20.11)	37 467	36 785	39.59 (39.19, 39.99)	38.87 (38.47, 39.27)
Aortic valve stenosis	226	210	1.27 (1.12, 1.45)	1.18 (1.03, 1.36)	1843	1767	1.95 (1.86, 2.04)	1.87 (1.78, 1.96)
Atrioventricular septal defect	700	678	3.95 (3.67, 4.25)	3.82 (3.55, 4.12)	7409	7143	7.83 (7.65, 8.01)	7.55 (7.37, 7.72)
Coarctation of aorta	1204	1177	6.79 (6.42, 7.18)	6.64 (6.27, 7.03)	9175	8832	9.69 (9.50, 9.89)	9.33 (9.14, 9.53)
Common truncus (truncus arteriosus)	127	126	0.72 (0.60, 0.85)	0.71 (0.60, 0.85)	1240	1213	1.31 (1.24, 1.39)	1.28 (1.21, 1.36)
Double outlet right ventricle (DORV)	405	395	2.28 (2.07, 2.52)	2.23 (2.02, 2.46)	4335	4103	4.58 (4.45, 4.72)	4.34 (4.20, 4.47)
Ebstein anomaly	110	109	0.62 (0.51, 0.75)	0.61 (0.51, 0.74)	1232	1195	1.30 (1.23, 1.38)	1.26 (1.19, 1.34)
Hypoplastic left heart syndrome	517	512	2.92 (2.67, 3.18)	2.89 (2.65, 3.15)	6430	6291	6.79 (6.63, 6.96)	6.65 (6.48, 6.81)
Interrupted aortic arch	109	106	0.61 (0.51, 0.74)	0.60 (0.49, 0.72)	962	907	1.02 (0.95, 1.08)	0.96 (0.90, 1.02)
Pulmonary valve atresia and stenosis	1688	1652	9.52 (9.08, 9.98)	9.32 (8.88, 9.78)	10 656	10 256	11.26 (11.05, 11.47)	10.84 (10.63, 11.05)
Pulmonary valve atresia	251	249	1.42 (1.25, 1.60)	1.40 (1.24, 1.59)	2260	2169	2.39 (2.29, 2.49)	2.29 (2.20, 2.39)
Single ventricle	223	215	1.26 (1.10, 1.43)	1.21 (1.06, 1.39)	1964	1883	2.08 (1.99, 2.17)	1.99 (1.90, 2.08)
Tetralogy of Fallot	856	850	4.83 (4.51, 5.16)	4.79 (4.48, 5.13)	9074	8936	9.59 (9.39, 9.79)	9.44 (9.25, 9.64)
Transposition of great arteries (TGA)	460	457	2.59 (2.37, 2.84)	2.58 (2.35, 2.82)	4566	4446	4.82 (4.69, 4.97)	4.70 (4.56, 4.84)
D-TGA	380	377	2.14 (1.94, 2.37)	2.13 (1.92, 2.35)	3517	3426	3.72 (3.60, 3.84)	3.62 (3.50, 3.74)
Tricuspid valve atresia and stenosis	194	193	1.09 (0.95, 1.26)	1.09 (0.95, 1.25)	2102	2036	2.22 (2.13, 2.32)	2.15 (2.06, 2.25)
Ventricular septal defect	10 452	10 105	58.94 (57.82, 60.08)	56.98 (55.88, 58.10)	58 065	55 614	61.35 (60.85, 61.85)	58.76 (58.27, 59.25)
Total anomalous pulmonary venous return (TAPVR)	161	157	0.91 (0.78, 1.06)	0.89 (0.76, 1.04)	1957	1911	2.07 (1.98, 2.16)	2.02 (1.93, 2.11)
Orofacial
Choanal atresia	275	238	1.55 (1.38, 1.75)	1.34 (1.18, 1.52)	2229	1987	2.36 (2.26, 2.45)	2.10 (2.01, 2.19)
Cleft lip with cleft palate	846	820	4.77 (4.46, 5.10)	4.62 (4.32, 4.95)	10 148	9691	10.72 (10.52, 10.93)	10.24 (10.04, 10.45)
Cleft lip without cleft palate	402	396	2.27 (2.06, 2.50)	2.23 (2.02, 2.46)	3781	3717	3.99 (3.87, 4.12)	3.93 (3.80, 4.06)
Cleft palate without cleft lip	935	860	5.27 (4.95, 5.62)	4.85 (4.54, 5.18)	9482	8744	10.02 (9.82, 10.22)	9.24 (9.05, 9.43)
Gastrointestinal
Biliary atresia	159	153	0.90 (0.77, 1.05)	0.86 (0.74, 1.01)	2325	2256	2.46 (2.36, 2.56)	2.38 (2.29, 2.48)
Esophageal atresia/ tracheoesophageal fistula	392	378	2.21 (2.00, 2.44)	2.13 (1.93, 2.36)	3918	3825	4.14 (4.01, 4.27)	4.04 (3.92, 4.17)
Rectal and large intestinal atresia/stenosis	746	712	4.21 (3.92, 4.52)	4.01 (3.73, 4.32)	8243	7990	8.71 (8.52, 8.90)	8.44 (8.26, 8.63)
Small intestinal atresia	894	833	5.04 (4.72, 5.38)	4.70 (4.39, 5.03)	5529	5278	5.84 (5.69, 6.00)	5.58 (5.43, 5.73)
Genitourinary
Bladder exstrophy	44	42	0.25 (0.18, 0.33)	0.24 (0.18, 0.32)	434	424	0.46 (0.42, 0.50)	0.45 (0.41, 0.49)
Hypospadias	5804	5513	63.93 (62.30, 65.59)	60.72 (59.14, 62.35)	31 135	29 450	63.57 (62.86, 64.28)	60.13 (59.44, 60.82)
Congenital posterior urethral valves	217	207	1.22 (1.07, 1.40)	1.17 (1.02, 1.34)	1392	1321	1.47 (1.40, 1.55)	1.40 (1.32, 1.47)
Renal agenesis/hypoplasia	785	767	4.43 (4.13, 4.75)	4.33 (4.03, 4.64)	5212	5064	5.51 (5.36, 5.66)	5.35 (5.21, 5.50)
Cloacal exstrophy	1110	1040	6.26 (5.90, 6.64)	5.86 (5.52, 6.23)	6990	6574	7.39 (7.21, 7.56)	6.95 (6.78, 7.12)
Musculoskeletal
Diaphragmatic hernia	538	524	3.03 (2.79, 3.30)	2.95 (2.71, 3.22)	4321	4191	4.57 (4.43, 4.70)	4.43 (4.30, 4.56)
Gastroschisis	819	805	4.62 (4.31, 4.95)	4.54 (4.24, 4.86)	4996	4874	5.28 (5.13, 5.43)	5.15 (5.01, 5.30)
Omphalocele	172	170	0.97 (0.84, 1.13)	0.96 (0.82, 1.11)	1230	1183	1.30 (1.23, 1.37)	1.25 (1.18, 1.32)
All limb deficiencies	614	531	3.46 (3.20, 3.75)	2.99 (2.75, 3.26)	3771	3350	3.98 (3.86, 4.11)	3.54 (3.42, 3.66)
Clubfoot	2207	2063	12.45 (11.94, 12.98)	11.63 (11.14, 12.15)	14 269	13 198	15.08 (14.83, 15.33)	13.94 (13.71, 14.18)
Chromosomal
Trisomy 13	135	127	0.76 (0.64, 0.90)	0.72 (0.60, 0.85)	980	942	1.04 (0.97, 1.10)	1.00 (0.93, 1.06)
Trisomy 18	300	294	1.69 (1.51, 1.89)	1.66 (1.48, 1.86)	2021	1946	2.14 (2.04, 2.23)	2.06 (1.97, 2.15)
Trisomy 21 (Down syndrome)	2249	2160	12.68 (12.17, 13.22)	12.18 (11.68, 12.70)	23 214	21 773	24.53 (24.21, 24.85)	23.01 (22.70, 23.31)
Deletion 22 q11	51	46	0.29 (0.22, 0.38)	0.26 (0.19, 0.35)	674	596	0.71 (0.66, 0.77)	0.63 (0.58, 0.68)
Turner syndrome	106	97	1.22 (1.01, 1.48)	1.12 (0.92, 1.37)	960	885	2.11 (1.98, 2.25)	1.94 (1.82, 2.08)

Bold text for reported rates indicates birth defects in which the prevalence estimated using only 10 codes fell outside of the confidence interval for the prevalence estimated using all available codes.

^aIncludes birth defects whose case counts and prevalence rates are published in Birth Defects Research as part of the annual data report from the National Birth Defects Prevention Network.

^bICD-9-CM codes listed on any hospitalization in which the date of admission was between 0 and 364 days after the child’s date of birth are included.

^cFrequencies and rates are not reported for birth defects with fewer than 10 total cases.

As shown in Table 2, we observed considerable variation in the proportion of cases that would have been missed across specific birth defects if fewer than the total number of available diagnosis code fields were used to identify cases. When restricting the FBDR to 10 diagnosis codes, less than 1% of cases among 8 birth defects, including 7 congenital heart defects and anencephaly, would have been missed, the latter of which would not have lost any of its 83 cases. Conversely, more than 10% of cases among 6 birth defects would have been missed, with anophthalmia/microphthalmia having the highest percentage of cases lost (22.2%) if only the first 10 codes were available. When restricting the HCUP KID to the first 10 diagnosis codes, only anencephaly would have missed <1% of its cases. More than 10% of cases would have been missed for 7 birth defects, with anophthalmia/microphthalmia again having the highest percentage of cases lost (20.7%).

Table 2.

Comparison of the proportion of cases of selected birth defects^athat would be missed if fewer than 31 diagnosis codes from inpatient hospitalization records^bare used to identify cases, Florida Birth Defects Registry, 2006-2013, and if fewer than 25 diagnosis codes from inpatient hospitalization records^bare used to identify cases, KID Inpatient Database, Healthcare Cost and Utilization Project, 2009, 2012

Category/defect	FBDR						HCUP KID
	Relative to using all 31 available ICD-9-CM diagnosis codes, % of cases missed if restriction was made to:						Relative to using all 25 available ICD-9-CM diagnosis codes, % of cases missed if restriction was made to:
	Codes 1-25	Codes 1-20	Codes 1-15	Codes 1-10	Codes 1-5	Principal onlyc	Codes 1-25	Codes 1-20	Codes 1-15	Codes 1-10	Codes 1-5	Principal onlyc
Central nervous system
Anencephalus	0.0	0.0	0.0	0.0	1.2	97.6	Ref	0.0	0.0	0.6	3.5	97.6
Encephalocele	0.0	0.0	0.7	2.7	10.7	69.1	Ref	0.0	0.5	2.1	12.4	74.5
Holoprosencephaly	0.0	0.4	2.3	5.0	15.3	95.1	Ref	0.1	0.6	5.4	17.3	95.6
Spina bifida without anencephalus	0.0	0.2	1.1	1.7	7.3	69.7	Ref	0.2	0.5	2.0	9.1	75.9
Eye
Aniridia	0.0	0.0	14.3	14.3	42.9	100.0	Ref	0.0	3.8	10.6	29.8	90.4
Anophthalmia/ microphthalmia	0.7	3.5	9.7	22.2	53.5	97.2	Ref	1.9	6.0	20.7	48.5	98.5
Congenital cataract	1.9	3.2	9.6	19.9	38.5	91.0	Ref	1.7	5.1	15.5	37.8	87.5
Ear
Anotia/microtia	0.8	1.6	6.4	21.6	47.2	100.0	Ref	1.8	5.1	15.7	41.2	99.5
Cardiovascular
Primary CCHD	0.0	0.0	0.2	0.8	3.3	47.0	Ref	0.0	0.2	1.9	6.9	56.8
Secondary CCHD	0.0	0.1	0.4	1.8	10.7	65.3	Ref	0.1	0.3	3.4	14.5	66.8
Primary or secondary CCHD	0.0	0.0	0.2	0.8	4.9	49.6	Ref	0.0	0.1	1.8	7.5	56.7
Aortic valve stenosis	0.0	2.2	3.1	7.1	23.9	78.3	Ref	0.4	1.2	4.1	21.1	73.1
Atrioventricular septal defect	0.1	0.1	0.4	3.1	11.1	52.3	Ref	0.1	0.4	3.6	16.0	66.0
Coarctation of aorta	0.0	0.1	0.4	2.2	14.8	68.3	Ref	0.1	0.3	3.7	18.3	66.7
Common truncus (truncus arteriosus)	0.0	0.0	0.0	0.8	5.5	63.0	Ref	0.0	0.3	2.2	10.1	62.6
Double outlet right ventricle (DORV)	0.0	0.0	0.7	2.5	9.1	61.0	Ref	0.1	0.6	5.4	17.0	70.2
Ebstein anomaly	0.0	0.0	0.9	0.9	4.5	82.7	Ref	0.0	0.1	3.0	11.8	78.2
Hypoplastic left heart syndrome	0.0	0.0	0.4	1.0	2.3	48.2	Ref	0.0	0.2	2.2	6.6	60.2
Interrupted aortic arch	0.0	0.9	0.9	2.8	8.3	64.2	Ref	0.0	0.3	5.7	16.0	67.8
Pulmonary valve atresia and stenosis	0.1	0.2	0.5	2.1	13.7	88.3	Ref	0.2	0.6	3.8	19.6	86.3
Pulmonary valve atresia	0.0	0.0	0.0	0.8	5.6	74.9	Ref	0.2	0.5	4.0	12.1	77.7
Single ventricle	0.0	0.9	1.3	3.6	12.1	75.8	Ref	0.2	0.8	4.1	16.9	75.4
Tetralogy of Fallot	0.0	0.0	0.2	0.7	3.9	42.3	Ref	0.0	0.1	1.5	5.6	52.9
Transposition of great arteries (TGA)	0.0	0.0	0.0	0.7	5.7	51.3	Ref	0.2	0.4	2.6	11.3	62.1
D-TGA	0.0	0.0	0.0	0.8	4.7	51.8	Ref	0.2	0.4	2.6	10.0	59.7
Tricuspid valve atresia and stenosis	0.0	0.0	0.0	0.5	4.6	68.0	Ref	0.0	0.3	3.1	12.9	70.9
Ventricular septal defect	0.0	0.2	0.6	3.3	13.8	94.8	Ref	0.1	0.6	4.2	21.3	92.4
Total anomalous pulmonary venous return (TAPVR)	0.0	0.0	0.0	2.5	12.4	50.9	Ref	0.0	0.1	2.4	14.1	48.9
Orofacial
Choanal atresia	0.7	2.5	6.2	13.5	35.3	82.5	Ref	0.6	3.5	10.9	30.6	82.1
Cleft lip with cleft palate	0.1	0.2	0.5	3.1	12.3	57.2	Ref	0.2	1.1	4.5	16.1	64.6
Cleft lip without cleft palate	0.0	0.2	0.7	1.5	9.7	62.2	Ref	0.1	0.3	1.7	8.8	67.7
Cleft palate without cleft lip	0.3	1.6	3.2	8.0	20.5	67.1	Ref	0.6	2.0	7.8	29.3	80.1
Gastrointestinal
Biliary atresia	0.0	0.6	1.9	3.8	13.2	48.4	Ref	0.0	0.1	3.0	10.8	60.4
Esophageal atresia/ tracheoesophageal fistula	0.5	0.5	1.3	3.6	15.6	75.0	Ref	0.1	0.2	2.4	10.6	66.7
Rectal and large intestinal atresia/stenosis	0.1	0.1	0.8	4.6	16.8	63.1	Ref	0.1	0.4	3.1	15.3	63.6
Small intestinal atresia	0.2	0.7	1.9	6.8	24.4	82.8	Ref	0.2	0.9	4.5	19.9	74.7
Genitourinary
Bladder exstrophy	0.0	0.0	0.0	4.5	15.9	61.4	Ref	0.0	0.0	2.3	8.8	60.4
Hypospadias	0.3	0.7	1.6	5.0	17.3	99.4	Ref	0.4	1.5	5.4	19.3	99.1
Congenital posterior urethral valves	0.0	0.5	1.8	4.6	18.9	79.7	Ref	0.0	0.6	5.1	21.9	74.0
Renal agenesis/hypoplasia	0.1	0.5	1.0	2.3	10.2	98.5	Ref	0.1	0.5	2.8	13.1	99.0
Cloacal exstrophy	0.1	0.4	1.2	6.3	21.3	87.6	Ref	0.3	1.2	6.0	23.1	83.5
Musculoskeletal
Diaphragmatic hernia	0.0	0.2	0.7	2.6	12.1	77.1	Ref	0.0	0.1	3.0	10.1	68.9
Gastroschisis	0.0	0.2	0.5	1.7	7.8	83.0	Ref	0.1	0.1	2.4	8.5	76.9
Omphalocele	0.0	0.0	0.6	1.2	7.6	80.8	Ref	0.0	0.2	3.8	11.7	78.0
All limb deficiencies	0.5	2.4	6.7	13.5	30.5	97.7	Ref	0.7	3.8	11.2	32.5	98.1
Clubfoot	0.2	0.9	2.4	6.5	20.2	97.8	Ref	0.4	2.1	7.5	25.2	97.4
Chromosomal
Trisomy 13	0.7	0.7	2.2	5.9	15.6	94.8	Ref	0.0	0.2	3.9	14.6	94.2
Trisomy 18	0.0	0.7	1.0	2.0	12.7	94.0	Ref	0.0	0.2	3.7	16.1	94.7
Trisomy 21 (Down syndrome)	0.3	0.6	1.4	4.0	21.6	99.1	Ref	0.3	1.4	6.2	30.8	98.7
Deletion 22 q11	2.0	3.9	3.9	9.8	25.5	98.0	Ref	0.4	1.6	11.6	33.5	99.0
Turner syndrome	0.0	0.9	3.8	8.5	25.5	97.2	Ref	0.5	1.7	7.8	34.6	96.9

^aIncludes birth defects whose case counts and prevalence rates are published in Birth Defects Research as part of the annual data report from the National Birth Defects Prevention Network.

^bICD-9-CM codes listed on any hospitalization in which the date of admission was between 0 and 364 days after the child’s date of birth are included.

^cThe extremely high proportion of cases missed if restriction was made to the principal diagnosis code field only, particularly for birth defects diagnosed at birth and with smaller likelihood of readmission throughout the first year of life, is due to the common practice of documenting a V30-39 code (indicative of the birth itself) as the principal diagnosis for birth hospitalizations. As such, for nearly all birth hospitalizations, the birth defect will not be documented as the principal diagnosis.

As expected, in both the FBDR and KID, the proportion of cases that would have been missed increased as the number of code fields used to identify cases was decreased. In general, using 25 code fields resulted in a minimal loss of cases overall (0.2%), with no defects missing >4% of cases that would have been captured using all 31 code fields for the FBDR. Using 20 code fields in the KID resulted in a loss of less than 2% for all birth defects compared to the full data with 25 code fields. Restriction to 5 codes would have resulted in failure to capture 10.9% of all birth defect cases, and nearly 84% of cases would have been missed if only the principal diagnosis code were used in the FBDR, with similar findings in the KID. The reason for the extremely high proportion of cases missed when restricting to the principal diagnosis code is due to the common practice of documenting a V30-39 ICD-9-CM code (indicative of the birth) as the principal diagnosis for birth hospitalizations. As such, for defects often diagnosed at birth and for which a repeat defect-related hospitalization in the first year of life is rare, the birth hospitalization is the only inpatient record on which the defect is captured, and since the defect will not be documented as the principal diagnosis on the birth hospitalization, a high proportion of cases would be missed.

The impact of the number of diagnosis code fields used in the FBDR on the isolated vs multiple classification of birth defects is shown in Figure 1. Among 6 birth defects, the proportion of cases that was classified as isolated increased 5% or more when only 10 codes were used, due to the inability to capture the co-occurring defect that would have resulted in a “multiple” classification. For example, when 31 codes were used, 52.8% of the 125 anotia/microtia cases identified were classified as isolated (having a body part/system other than the ear impacted by a birth defect). However, when restricting to 10 codes, not only were fewer cases diagnosed (98), but over 64% were classified as being isolated due to the inability to capture other defects appearing in code fields 11 to 31.

Figure 1.

Assessing the impact of the number of inpatient ICD-9-CM discharge diagnosis codes^a used on the classification of selected birth defects^b as isolated vs multiple^c, Florida Birth Defects Registry, 2006-2013. ^aICD-9-CM codes listed on any hospitalization in which the date of admission was between 0 and 364 days after the child’s date of birth are included. ^bIncludes birth defects whose case counts and prevalence rates are published annually in Birth Defects Research: Part A. ^cTo be considered a “multiple” birth defect, the infant must have been diagnosed with the listed birth defect and another birth defect from a different body system. Infants with multiple defects from the same body system (eg, trisomy 21 and trisomy 18), but not another body system would be classified as “isolated” in this figure. Note: Numbers appearing to the right of each bar represent the additional percentage of cases identified as “isolated” following restriction to only 10 diagnosis codes per discharge record. For 2 defects (Ebstein anomaly and cleft lip without cleft palate), the proportion of isolated cases decreased with code restriction, as fewer cases were identified, and the proportion of those identified that were isolated was lower.

Supplementary Table S1 provides a listing of all ICD-9-CM birth defect codes from the FBDR that were documented on at least 100 inpatient discharge records from 2006 to 2013, stratified by the proportion of time the code was listed outside the first 10 diagnoses. The most common birth defect codes included 745.5 (atrial septal defect, n = 56 423), 757.33 (congenital pigmentary anomalies of the skin, n = 43 430), and 747.0 (patent ductus arteriosus, n = 36 002). There were 15 diagnoses that were documented outside the first 10 codes in >25% of the discharge records in which the code was listed, with the 550.92 (non-recurrent bilateral inguinal hernia without mention of obstruction or gangrene) and 748.60 (unspecified anomaly of the lung) codes appearing outside the first 10 codes more than half of the time. There were 26 codes outside the first 10 codes 15% to 25% of the time, 94 outside the first 10 codes 5% to 15% of the time, and 77 outside the first 10 codes less than 5% of the time.

Supplementary Table S2 provides a national snapshot of comparable KID information in 2009 and 2012. There were 54 individual codes that were documented outside the first 10 codes in over 25% of the discharge records in which the code was listed, with no birth defect codes appearing outside the first 10 codes more than half of the time. There were 34 codes outside the first 10 codes 15% to 25% of the time, 86 outside the first 10 codes 5% to 15% of the time, and 46 outside the first 10 codes less than 5% of the time. The codes that were listed outside the first 10 codes <1% of the time in both the FBDR and KID databases tended to be serious, easily recognizable defects such as 746.7 (hypoplastic left heart syndrome), 745.2 (tetralogy of Fallot), 745.10 (transposition of the great arteries), and 741.03 (spina bifida).

DISCUSSION

Evaluation studies have been valuable in providing information germane to the interpretation of birth defects surveillance data. In particular, such evaluations can provide explanations for variation in the prevalence of birth defects over time other than changes in the true underlying prevalence. The FBDR has investigated the independent and joint contribution of a changing case ascertainment net (ie, differing data sources) over time,⁷ the temporal changes in completeness of ascertainment,² and the variation in accuracy of diagnosis codes for identifying birth defects.¹⁸ This study demonstrates that the number of ICD-9-CM code fields available in an inpatient database to identify infants with birth defects can have a substantive impact on a surveillance program’s completeness, its reported prevalence estimates, and its ability to accurately classify defects as isolated or multiple. Studies seeking to leverage nationally representative cross-sectional databases such as the HCUP-KID to estimate prevalence are similarly susceptible to the number of diagnosis code fields available to identify hospitalizations indicative of birth defects. Moreover, the anticipated impact varies by both the number of fields available as well as the nature of the birth defect(s) of interest.

In Florida in 2006, the number of ICD-9-CM diagnosis code fields tripled, from 10 to 31. In 2009, the KID database, and the similar nationally representative, cross-sectional, all-payer database for inpatients of all ages, the National Inpatient Sample (NIS), increased from 15 to 25 code fields. For birth defects diagnoses, the expansion prevented the loss of 1 in 40 cases that would not have been noted if only the first 10 codes were available. The impact of expansion beyond 10 diagnosis code fields was minimal for severe (eg, hypoplastic left heart syndrome), lethal (eg, anencephaly), or easily recognized (eg, gastroschisis) defects. Although neither diagnostic coding processes in hospitals nor their variations throughout Florida or nationally are fully understood, it is likely that such life-threatening or visually striking conditions are routinely coded as the principal or first few secondary diagnoses and therefore are more likely to be captured, even if a dataset contains only a small number of diagnosis code fields.

Previous research has investigated the extent to which the number of diagnosis code fields available affects prevalence and reporting of other conditions. Dubberke et al.¹⁰ conducted a study in 4 US hospitals and reported that, compared to using only the first 7 ICD-9-CM diagnosis codes, the reported prevalence of Clostridium difficile infection using the first 15 codes increased by 12%. However, the impact of extending to all documented codes was minimal and not statistically significantly different from using 15 codes. Drosler et al.,⁹ representing the World Health Organization Quality and Safety Topic Advisory Group, analyzed data from 6 countries and 2 US states (California and Florida) to explore the optimal number of diagnosis fields in administrative data needed to capture patient safety indicators (eg, postoperative sepsis). The authors concluded that systems with fewer than 15 secondary diagnosis fields are “likely to lose relevant clinical information.” Increasing the number of diagnosis code fields has also been associated with higher prevalence of chronic comorbidities, such as hypertension, diabetes, obesity, depression, and cancer,¹²,¹³ and restriction to fewer than 10 codes, compared to 25, has even been shown to result in biased associations between these conditions and mortality.¹¹,¹³

An understanding of how the number of diagnostic codes available impacts the ability to identify cases, and the variability of that impact by condition subtype and overall case severity, is extremely valuable to researchers and public health planners. This is because even when national databases such as the NIS make standardized changes to the publicly available files (eg, increase from 25 to 30 code fields), 60% of states/jurisdictions reporting to the NIS in 2014 (27 of 45) themselves reported fewer than 30 diagnosis code fields, and 3 reported fewer than 10 fields. Although the sampling designs of the NIS and KID do not support state-level analyses, an understanding of the potential effects of these variations on descriptive or analytic studies is important.¹⁹ For example, following the emergence of Zika virus as a 2016 Public Health Emergency of International Concern, some surveillance programs relied on administrative data in an attempt to quickly establish estimates of the temporal trends of microcephaly in the “pre-Zika” era. In our study, the 742.1 code for microcephaly was listed outside the top 10 codes 20% to 25% of the time; therefore, any changes in the number of diagnosis code fields available could have confounded the trends assessment.

The implications of our findings also translate to other clinical conditions identified by diagnostic codes and in which the number of available fields is less than the total number of diagnoses made for any given patient. The magnitude of the impact, however, likely depends on the specific condition, number of coding fields, and diagnostic coding practices, underscoring the need for further investigation.

A few limitations should be considered when interpreting the results of this study. The FBDR is a passive surveillance system that relies on the linkage of administrative data sets and ICD-based codes to identify birth defect cases; these diagnoses are not verified through medical record review or diagnostic confirmation. Inaccuracies in ICD codes have been documented in the literature and may result in false positive diagnoses;¹⁸ however, we have little evidence that accuracy changed significantly during the study time frame and were responsible for the observed temporal changes. The FBDR does not consider diagnoses documented during encounters after the child’s first birthday, and although the majority of reported birth defects are diagnosed at birth or within the first few months of life, there is a likelihood for some missed cases, especially for birth defects that are less severe and manifest later in life. Last, we were unable to investigate, neither within the state-level FBDR nor the nationally representative KID, variation in coding practices—protocols for documenting, storing, ordering, and disseminating diagnostic coding information in inpatient discharge records—within and across the thousands of hospitals whose data were included in this evaluation. More research is needed to elucidate these levels of (in)consistency in activities that lead to the production and dissemination of these discharge databases, particularly for complex cases with a great deal of diagnostic information in the medical record.

This study assesses the impact of the expansion of diagnosis codes on frequencies and rates of major birth defects, and on the classification of ascertained defects as isolated or multiple. The large sample sizes at the state and national levels are particularly important for uncommon health outcomes, and facilitated the investigation of a number of rare individual birth defects, ranging in severity, rather than more heterogeneous groupings of defects based on body system. Second, our study not only included inpatient databases representing a state-level registry and national-level sample, but that also varied in their design and use. Although the same hospitalization-level analyses are undertaken, the FBDR is a longitudinally linked registry that results in an unduplicated inventory of a birth cohort of children who are diagnosed with included birth defects within the first year of life. As such, prevalence estimates from the FBDR reflect cases per 10 000 live births. In contrast, the KID is a hospitalization-level, cross-sectional database in which the user is unable to distinguish whether 2 hospitalizations are for the same or different infants (ie, infants with multiple hospitalizations indicative of a birth defect are counted multiple times). This, in part, explains the markedly higher prevalence rates of birth defects using the KID vs FBDR, which also has implications for comparisons of epidemiological or economic investigations of birth defects using various types of databases. Including both sources of inpatient diagnoses among children less than 1 year of age demonstrates the relative consistency of the impact of changing numbers of diagnosis code fields, regardless of sampling design or database construction.

CONCLUSION

Despite collaborative prevention, education, and research efforts nationally and internationally, birth defects remain common, costly, and critical.^20–22 Data from the FBDR and other surveillance programs worldwide are used extensively to inform epidemiologic and clinical research, public policy, prevention and education programs, and myriad other activities.³ Our findings encourage additional research for other health outcomes in patients of all ages. Other disease registries, too, whether using active or passive ascertainment protocols, rely at least in part on diagnostic codes documented by healthcare providers in their case-finding activities, making routine investigation of these databases essential. Moreover, large publicly available hospital discharge or claims databases used frequently by researchers to investigate temporal trends, risk factors, clinical outcomes, or healthcare utilization should be held to the same level of ongoing examination.

FUNDING

This work was supported by the Centers for Disease Control and Prevention grant number 1 NU50DD004946-01-00.

CONTRIBUTORS

JS and JT were responsible for the evaluation framework and study design. JS, RR, and JT were responsible for compiling, cleaning, and aggregating the data. JS and JT were responsible for analyzing the state data; JS and JM were responsible for analyzing the national data. All authors were involved in substantive interpretation of all data analysis, including development of table shells and conceptualization of figures. All of the coauthors participated in drafting of the manuscript and revising it critically for important intellectual content, and all have provided their approval of the final version to be published. All coauthors agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

SUPPLEMENTARY MATERIAL

Supplementary material is available at Journal of the American Medical Informatics Association online.

Conflict of interest statement. The authors have no competing interests to declare.