Archives of Medical Research 34 (2003) 214221

ORIGINAL ARTICLE
Discriminant Function of Perinatal Risk That Predicts Early Neonatal
Morbidity: Its Validity and Reliability
Rita Esther Zapata-Vazquez,
a,b
Luis Alfonso Rodr guez-Carvajal,
d
Gilberto Sierra-Basto,
e
Felipe Manuel Alonzo-Vazquez
b
and Manuel Echeverr a-Egu luz
c
a
Unidad Pediatrica de Terapia Intensiva,
b
Departamentos de Pediatr a y Ginecolog a,
c
Departamentos de Neonatolog a, Pediatr a y Ginecolog a,
Hospital Carlos Urzaiz Jimenez, Instituto Mexicano del Seguro Social (IMSS), Merida, Yucatan, Mexico
d
Facultad de Matematicas, Universidad Autonoma de Yucatan (UADY), Merida, Yucatan, Mexico
e
Unidad Pediatrica de Terapia Intensiva, Hospital Benito Juarez, IMSS, Merida, Yucatan, Mexico
Received for publication April 26, 2002; accepted December 13, 2002 (02/086).
Background. This study aimed to identify signicant perinatal risk factors associated with
neonatal morbidity to construct a scoring system to aid in distinguishing between healthy
and ill neonates. Validity and reliability of the scoring system were determined.
Methods. We conducted a screening test and used logistic regression to analyze data from
a cohort of 387neonates andtodetermine therelationshipbetweenriskfactors andmorbidity.
Twenty nine factors of perinatal risk were studied. Logistic regression and discriminant
analysis were performed to assess risk for morbidity. This system was developed and
validated prospectively on 238 new neonates.
Results. Risk factors that demonstrated association with morbidity by logistic regression
were chronic maternal illness, premature rupture of membranes (PROM), amniotic uid,
low Apgar score at 5 min, obstetric trauma, hypertension, neonatal resuscitation, breathing
pattern at 6 h after delivery, birth weight, and gestational age. Discriminant function
obtained from discriminant analysis had sensitivity of 68% and specicity of 93%, while
positive and negative predictive values were 88 and 86%, respectively. Area belowreceiver
operating characteristic (ROC) curve was 0.86 (standard error [SE]: 0.02). In the validity
study, these values were maintained without signicant differences. Kappa statistic between
two physicians was calculated at 0.84 ( p 0.001).
Conclusions. Evidence indicated that discriminant function is a useful tool to assess ini-
tial neonatal risk, allowing pediatricians to predict morbidity prior to discharge of neonates.

2003 IMSS. Published by Elsevier Science Inc.

Key Words: Perinatal risk factors, Patient outcome assessment, Neonatal screening, Morbidity,
Patient discharge, Discriminant analysis, Logistic regression, Reproducibility of results.
Introduction
Clinicians frequently estimate likely prognosis for neonate
within the rst days of life. Subsequently, an actual dilemma
usually exists for those requiring discharge from a public
hospital. We have posed two essential questions: (1) What
Adress reprint requests to: Rita Esther Zapata-Vazquez, Calle 23-A
#308 entre 8 y 10, Fracc. San Esteban, 97149 Merida, Yucatan, Mexico.
Phone: (52) (999) 924-3003; FAX: (52) (999) 924-0554; E-mail:
zvazquez@tunku.uady.mx
0188-4409/03 $see front matter. Copyright

2003 IMSS. Published by Elsevier Science Inc.

doi : 10. 1016/ S0188- 4409( 03) 00029- 8
is the probability that this neonate will be unhealthy?, and
(2) If this baby remains at the hospital, what is the probabil-
ity that the stay for many additional hours will not be neces-
sary? These questions usually present themselves when
physicians talk about prognosis with parents and when physi-
cians consider discharge 6 h after delivery, because morbid-
ity is higher during the rst hours postpartum and the
following 7 days (1).
Currently, clinicians amalgamate prognostic data from a
variety of sources when predicting outcome for any given
neonate, but amount of risk factors previously studied
Discriminant Function That Predicts Neonatal Morbidity 215
overrides clinical capacity to assign to each factor the impor-
tance required for individual neonates. This study was
carried out to produce a tool to make this decision easier
based on risk factors present in each neonate. Early neonatal
morbidity in the present study related to morbidity during
the initial 7 days of postpartum due to maternal pathology
or to acute or chronic hypoxia.
Discriminant function referred to a function that assigns
scores to a selected set of factors that determines illness risk.
Use of scores at present is very helpful when clinical deci-
sions are to be made. Many institutions have implemented
these routinely for use in predicting morbidity and mortal-
ity (2,3).
Presently, a number of neonatal risk scores have been
developed to assure more accurate evaluation concerning
neonatal prognosis. There are many factors that contribute
to neonatal morbidity (4,5): hypertension, preterm labor (6),
chorioamnionitis, premature delivery, premature rupture of
membranes (PROM) of 12 h or more, endometritis (7),
and presence of meconium(8). Some authors have combined
several risk factors, e.g., PROMwith fever and foul-smelling
maternal ow (9). Perlman suggested that intubation, low
pH in umbilical cord, and low Apgar score at 5 min can
predict morbidity (10).
Prognosis of morbidity by means of scores that include
several variables were described by Portman (11) and Hobel
(12). Lubchenco developed a neonatal morbidity model (13).
In Mexico, there is no scoring system to determine
whether a neonate will continue to be healthy or not. In
addition, even when most risk factors are universal it is im-
portant to know which risk factors have a greater inuence
on morbidity.
Principal risk factors reported in medical practice were
studied according to frequency or severity, and one discrimi-
nant function was carefully designed to help physicians
decide whether or not neonates could be discharged early.
Reproducibility of results (validity and reliability) was
checked and reported.
Materials and Methods
Our work included alive, premature, full-term, or post-term
neonates. Data were collected from November 24 through
December 31, 1997 at the Carlos Urziaz Jimenez Hospital in
Merida, Yucatan, Mexico (a top-level federal social security
[IMSS] hospital). The study did not include patients with
severe congenital malformations. Size of sample group (458
neonates) was estimated on the basis of the 1996 neonatal
morbidity report (515 of 5,425 neonates). It was calculated
based on simple random sample for proportions. Neonates
were studied according to order of arrival.
The study design included a screening test, and the neo-
nate cohort was followed up fromdelivery through the initial
7 days of life. All mothers were asked to attend checkup
sessions for neonates. A pediatrician performed a physical
examination and conducted personal interviews for the rst
7 days. In this way, we corroborated whether or not neonates
were healthy; the pediatrician did not know neonatal and
maternal clinical histories. Neonates whose mothers did not
attend review sessions at the hospital were visited at home.
A sick neonate was dened as a neonate with evidence
of pathology determined by physical examination, labora-
tory, and X-ray during the rst 7 days independently of the
neonates stay at the hospital or not. A healthy neonate was
dened as breast- or bottle-fed with normal neurologic and
physical examination, normal bowel emptying, and normal
vital signs and temperature during the rst 7 days even
when the neonate was kept at the hospital unnecessarily. The
neonate may have had some minor congenital malformations
such as hydrocele and/or physiologic jaundice.
Twenty nine main perinatal risk factors reported in medi-
cal practice were studied according to frequency or severity.
These risk factors included the following: maternal age;
prenatal control; parity; number of miscarriages; previous
neonatal pathologies; previous cesarean section; hyperten-
sion, abortion threat; pretermbirth threat; infection of urinary
ducts; vaginal spotting; chronic illness; conduction or induc-
tion labor; intrauterine tachycardia or bradycardia; abrupt
placenta; prolapsed cord; presence of meconium; PROM for
24 h, or 12 h if preceded by multiple vaginal examina-
tions; delivery care; Apgar score at 1 and 5 min; resusci-
tation; breathing pattern at 6 h after delivery; gestational
age; weight; height; gender; obstetric trauma, and blood
group no abortion (ABO) or Rh incompatibility. Multiple
logistic regression (14,15) was used to determine which
variables were signicant. Finally, discriminant analysis (16)
was applied to classify neonates into two groups and to
determine linear combination to differentiate them. Analyses
were made with SPSS, version 6 (SPSS, Inc., Chicago, IL,
USA). For validity and reliability of prognosis test, a new
cohort of neonates with the same characteristics as those in
the previous work was studied from May 23 through June
17, 1999.
Sample group size (263 neonates) was estimated based
on a neonatal morbidity report from a previous study as
18%. Apediatrician applied morbidity predictive score at the
time of birth. Six hours later, a second pediatrician blinded
to these data applied a second test using same predictive
score for all neonates. True morbidity was corroborated
during the rst 7 days and included physical examinations
and an inquiry, similar to the previous study. The study
was carried out under normal conditions; indications to ll
out discriminant function are shown in Appendix 1.
Prevalence, accuracy, sensitivity, specicity, positive and
negative predictive values, and area under receiver operating
characteristic (ROC) curve were calculated. Interobserver
variation was calculated by Kappa coefcient.
Zapata-Vazquez et al. / Archives of Medical Research 34 (2003) 214221 216
Results
A group of 458 neonates formed the basis of the study.
Seventy one neonates were discarded from the study because
it was not possible toconduct complete follow-up; of these, 11
were ill, 35 were healthy, and in the remaining 25 it was
not possible to corroborate neonate progress.
Neonates studied consisted of 71 sick (18.3%) and 316
(81.7%) healthy babies. Mean weight for sick and healthy
neonates was 3.55 and 2.77 kg, respectively, and standard
deviations (SDs) 0.45 and 0.79, respectively. There was
signicant difference (0.78) between these means (p
0.001) with 95% condence interval (95% CI) from 0.64
to 0.92. Mean gestational age for sick and healthy neonates
was 39 (SD 1.65) and 37 weeks (SD 3.16), respectively,
with signicant difference (p 0.001) and 95% CI 1.49 to
2.5. Chi-square independence test analysis of gender was
not signicant (2.23, p 0.10). Risk factors with signicant
results in logistic regression analysis were chronic mater-
nal illness (p 0.001), Apgar score at 5 min (p 0.001),
PROM (p 0.006), and the result of multiplying birth
weight gestational age (p 0.001); this product had better
t than each variable independently, amniotic uid (p 0.02),
and obstetric trauma (p 0.03).
Risk factors with odds ratio (OR) 1 in logistic regres-
sion were maternal hypertension (OR 1.5; 95% CI 0.6,
3.6), resuscitation, and breathing pattern at 6 h after delivery;
nonetheless, because these variables are polythomous,
they produced one OR for each outcome in comparison to
reference. Thus, four ORvalues can be seen for resuscitation,
including the following: 1) between oxygen and good breath-
ing (OR 1.13; 95% CI 0.38, 3.3); 2) between positive
pressure and good breathing (OR 6; 95% CI 1.03, 35); 3)
between tracheal intubation and good breathing (OR 3.6;
95% CI 0.07, 181), and 4) between drugs or cardiac massage
and good breathing (OR 3.4; 95% CI 1.88, 6.26). For
breathing pattern, two OR values may be seen: 1) between
bradypnea or tachypnea and normal pattern (OR 1.08;
95%CI 0.2, 5.9), and 2) between apnea or automated absence
and normal pattern (OR 3.4; 95% CI 0.28, 47).
These 10 risk factors were analyzed using discriminant
analysis. Birth weight and gestational age were entered as
isolated. By means of Wilks lambda statistics, we again
corroborated that variables in mathematical model had sta-
tistical signicance (Table 1).
Using unstandardized canonical discriminant function co-
efcients, a linear equation was constructed as follows:
Discriminant function 1.1834 1.33 (chronic ill-
ness) 1.92 (PROM) 0.44 (breathing pattern) 0.60
(obstetric trauma) 0.59 (amniotic uid) 0.16 (resuscita-
tion) 201.41 (1/gestational age) 1.86 (1/square root
of weight) 0.27 (hypertension) 0.91 (Apgar score at
5 min).
Certainty of discriminant function was evaluated using
receiver operating curve (ROC) (17). Figure 1 shows the
Table 1. Wilks lambda and univariate F-ratio with 1 and 386
of freedom
Risk factor Wilks lambda F Signicance (p)
a
Chronic maternal illness 0.977 9.06 0.003
Amniotic uid 0.967 13.32 0.000
Hypertension 0.989 4.26 0.039
PROM 0.971 11.60 0.000
Resuscitation 0.879 52.94 0.000
Breathing pattern 0.880 52.41 0.000
Gestational age 0.931 28.76 0.000
Apgar score at 5 min 0.823 82.58 0.000
Birth weight 0.936 26.52 0.000
Obstetric trauma 0.931 28.76 0.000
a
Signicance (p) level of differences between low and high risk.
relationship between sensitivity and specicity across all cut-
points of the test; area under curve estimated from maximum
likelihood was 0.86 (SE: 0.02). 95% CI ranged from 0.80
to 0.94. To separate neonates in the group who were sick
from those who were healthy, a cut-off level of discriminant
function must be set so that that above it neonates are desig-
nated as high-risk and under it as low-risk. Cut-off value
level was 0.7; this referred to highest accuracy (minimal
false negative and false positive results). A total of 88% of
neonates at this point classied correctly; sensitivity, speci-
city, and positive and negative predictive values are shown
in Table 2, while distribution of sick and healthy neonates
with discriminant function values is shown in Figure 2.
Appendix 2 shows discriminant function to be applied to
neonates at 6 h after delivery. Gestational age and birth
weight were calculated as a result of multiplying mid-point
of each class interval coefcient of discriminant function.
Score given 3.75 kg was to adjust to risk in clinical prac-
tice. Constant coefcient was included in coefcient to sim-
plify reduction operations. To render discriminant function
easier to handle, all scores were multiplied by 10 and rounded
off. This discriminant function does not cover the entire
Figure 1. Receiver operating characteristic (ROC) curve for discrimi-
nant function.
Discriminant Function That Predicts Neonatal Morbidity 217
Table 2. Classication of screening test results
Screening test to identify neonates with morbidity
True characteristics in neonates
Results of Predictive
screening Disease Healthy Total value
High 48 23 71 Positive
risk 7 () 68%
Low 23 293 316 Negative
risk 7 () 93%
Total 71 316 387
Sensitivity Specicity
68% 93%
spectrum of original variables, showing only birth weight
and gestational age of neonates who can be discharged early.
Causes of illness were prematurity, small for gestational
age, conjunctivitis, transient tachypnea of the newborn
(TTN), cephalhematoma, cardiac disease, abnormal jaundice,
patent ductus arteriosus, asphyxia, hypoxic-ischemic enceph-
alopathy (HIE), intracranial hemorrhage pulmonary adapta-
tion syndrome, respiratory distress syndrome (RDS), ABO
incompatibility, hyperkalemia and hypoglycemia, impetigo,
pneumonia, and sepsis, all classied as high risk.
There were three readmissions, including one patient with
pathologic jaundice, another showing poor weight gain and
sepsis, and the remaining patient, meningoencephalitis; the
latter neonate died during the rst 7 days after delivery. In
387 healthy neonates, only 21 presented one of the following:
caput succedaneum; hydrocele; preauricular cyst; cyst on
gums and lower incisors, and open rear fontanel; however,
these were not related to studied risk factors and subse-
quently did not put these neonates in danger if discharged
within 6 h.
Results of validity and reliability included 238 neonates
(90%) of the original 263 examined. Twenty ve (10%)
neonates were not considered in the study because we could
not locate them to complete follow-up.
Figure 2. Distribution on discriminant scores: high curve describes 316 healthy neonates, low curve describes 71 sick neonates.
Overall accuracy or concordance rate between estimated
scores and those of gold standard was 86%; sensitivity
was 64%, specicity was 94%, and positive and negative
predictives were 78 and 88%, respectively. Certainty was
86% for this study and 88% for the previous study. This gave
a nonsignicant difference (value 0.466); 95% CI for
difference between proportions was (0.033, 0.073).
Agreement between the two observers was assessed with
74 neonates by Kappa coefcient; we transformed results
into low- and high-risk, and value was 0.84 (p 0.001).
Pathologies found in sick neonate group with high-risk
scores were classied and shown in Appendix 3. It may be
observed that scores cover a wide range, i.e., from 7 to
values as high as 89.
Eleven patients were originally considered in false posi-
tive values; of these, six were born to mothers with PROM,
three were hospitalized and given treatment, but none
showed any clinical or pathologic signs or symptoms during
their stay at the hospital or at the end of the seventh day.
Other variables were bruises on the body, presence of meco-
niumuid, fetid amniotic uid, and small for gestational age.
Twenty two patients were considered false negatives. The
most common problem observed was infection such as om-
phalitis, otitis externa, neonatal sepsis, weight loss, hypogly-
cemia, cephalhematoma, pathologic jaundice. One clavicle
fracture and one TTN were not detected at 6 h postdelivery.
With regard to false negative results, the score did not
detect infection; nevertheless, infections developed between
the third and fourth day after birth. Therefore, there would
have been no reason to keep patients hospitalized. Weight-
decit neonates presented pathologic jaundice; mothers had
been instructed to breastfeed exclusively but did not have
sufcient milk.
Discussion and Analysis
In this screening test, 10 signs of risk of early neonatal
morbidity were found. Their sumprovided each newneonate
Zapata-Vazquez et al. / Archives of Medical Research 34 (2003) 214221 218
with a score, which allowed us to classify the neonates
into high- or low-risk groups; in addition, we were able to
determine whether the neonate could be discharged during
the rst 6 h of life with 88% certainty.
Birth weight and gestational age are good indicators
for risk evaluation. Nonetheless, two factors alone are in-
sufcient to predict morbidity or mortality with precision
(1821). Clinically, there are risk indicators reported as
unquestionable; they include maternal hypertension (22), ir-
regular breathing pattern, PROM (23), and pea-soup meco-
nium, which is irrefutably abnormal (8,24,25). It should be
mentioned that Apgar score at 1 min has little signicance as
opposed to Apgar score at 5 min, the latter of major impor-
tance in predicting morbidity (8,10,26).
Male neonates originally were considered at higher risk;
however, in this study we found no difference between
males and females. This nding was also reported earlier
(26,27). The most frequent causes of readmission reported
were physiologic jaundice (28), dehydration, dietary prob-
lems (29), apnea, and changes in breathing pattern. One
difculty encountered in our research was scarcity of similar
studies available for result comparison.
Most frequently used scores were developed from hospi-
tal les; however, les varied enormously in data accuracy
as well as in neonatal codication (30,31) during investiga-
tion of perinatal risk. The majority of researchers have used
les such as these (20) and determined their scores retrospec-
tively, using large samples and intending to arrive at valid
conclusions; examples can be found in References 27, 32,
and 33. Another difculty when comparing results of this
study with similar ones was the different types of statistical
analyses used, such as that of Singh (9), who employed
many 2 2 tables; this could bias signicance when making
a comparison.
One limitation of a study of this type is that in clinical
situations various factors occur in combination that are inter-
dependent, i.e., when conditions are not optimal other risk
factors tend to relate. A source of bias might be sample
size of some infrequent variables. Variables with OR1 were
included so that although these were not signicant, they
maintained a type II error as feasible.
The proposed model compared reasonably well with other
scores despite differences in sample size; for example, pedi-
atric risk of mortality has sensitivity of 67% and specicity
of 99% (34). Clinical risk index for babies (CRIB) had
sensitivity of 51% and specicity of 95% (3), and Tarnows
study had 81% sensitivity and 74% specicity (20). The
known score, Apache II, had sensitivity of 47% and specic-
ity of 95% (35).
Our study qualied with adequate sensitivity of 68% and
specicity of 93%. Probability of misclassication of the
proposed model was 12%; this meant that the study was
sufciently well aimed (88%) at the ability to diminish
number of unnecessary admissions and not allowing high-
risk patients to be discharged although they are apparently
healthy. Positive predictive value of 68% was inuenced
by low prevalence of sick neonates.
Test validity is dened as ability to distinguish between
neonates who had a disease and those who did not (36).
This was evaluated by ROC curve. It can also be interpreted
as follows: area of 0.86 meant that an individual randomly
selected from sick neonates had a test value larger than that
of a randomly chosen individual from healthy neonates 86%
of the time.
The perfect screening test required that all healthy neo-
nates have uniform value, and that sick neonates also have
uniform value, but a different value from that of healthy
neonates. This meant that results were consistent with those
of their respective groups (37). Factors that contribute to
variation for focusing on reliability (repeatability) included
interobserver variations measured by Kappa coefcient. Pro-
portion of neonates with morbidity was correctly identied in
reproducibility of results. These covered the entire spectrum
(healthy to sick neonates) of actual conditions at birth, as
in the initial sample.
There was no signicant difference between original dis-
criminant function score and validity and reliability concern-
ing certainty. We reported neonates with false positive as
well as false negative test results as a form of bias verica-
tion (38).
As with any risk estimator, pathology frequency was low;
consequently, positive predictive value of this score was not
as high as expected. Nonetheless, its high specicity made
us condent in its use.
The Committee of Fetus and Neonates at the American
Academy of Pediatrics stated that there is an element of
risk in early discharge of a patient (39). However, it also
recommended individualizing time of discharge according
to the medical report of the physician in turn, family eco-
nomic status, and social aspects of each case.
This did not contradict the protocol established for neo-
nates whose gestation was normal, full-term, and with clini-
cal assessment at a minimum of 6 h; this time should be
adjusted to provide adequate transitional period and a plan
for neonate feeding requirements (28). The fact that there
was low instance of morbidity (2%) (40) among full-term
neonates appeared to conrm that early discharge could be
feasible provided follow-ups are available for infants and
their mothers (31). Techniques of score measurement that
evaluate risk play an important role in research, resource
planning, and clinical decisions.
The purpose of discriminant function is to help the clini-
cian in daily practice; however, personal experience should
not be forgotten because of numeric score. The object of
the study focused on the neonates rst 7 days of life and
did not consider neonate evolution after this period. Risk
factors should be studied regionally with constant updat-
ing due to existing conditions that can affect incidence of
neonatal risk.
Discriminant Function That Predicts Neonatal Morbidity 219
Acknowledgments
We are very grateful to the neonatologists, in particular Drs. Nidia
Diego-Rodr guez and Martha Gomez de Vargas, at the IMSS Carlos
Urzaiz Jimenez Hospital Department of Pediatrics and Gynecology
in Merida, Yucatan, Mexico for their help in validating the model
for this work. In addition, we are also very grateful to Dr. Lorenzo
Osorno-Covarrubias for his support in carrying out this project.
The authors thank Luis Rodr guez-Carvajal, Ph.D., for statistical
support provided.
Appendix 1. Instructions for use of discriminant function
Circle scores, add scores together, and nally subtract
Apgar score to get discriminant function
1. Chronic maternal illness refers to any chronic disease:
cardiopathy, diabetes mellitus, rheumatoid arthritis,
herpes virus, cytomegalovirus, toxoplasmosis, hepati-
tis, HIV, etc.
2. Hypertension refers to increase in arterial pressure
(pregnancy-induced hypertension, preeclampsia,
eclampsia, and chronic hypertension).
3. Amniotic uid is examined using two aspects, the rst,
meconium uid or , and the second, pea
soup, fetid-smelling uid, an abnormal quantity of uid
that can be clinically classied as polyhydramnios
and oligohydramnios, or whether or not it con-
tains blood.
4. Gestational age was determined in weeks using Ballard
or Capurro method.
5. Resuscitation, taking into account when the neonate
needs oxygen, positive-pressure breathing with bag
and mask ventilation, tracheal intubation, cardiac mas-
sage, or drug therapy.
6. PROM is taken into consideration only if it lasts 24 h
or more, or 12 h and the mother has vaginal infection
or has undergone multiple vaginal examinations.
7. Obstetric trauma. A careful physical examination must
be performed to determine presence of caput, cephal-
hematoma, skin lesions (bruising), or fractures, espe-
cially broken clavicle. Mark only one.
8. Breathing pattern should be observed carefully to
verify frequency, detect bradypnea (35 per min),
tachypnea (60 per min), or automated absence.
9. Birth weight in kg.
10. Apgar score at 5 min must be subtracted from total
score of remaining parameters.
Appendix 2. Discriminant function of perinatal risk
that predicted morbidity
Date Time Name
Circle scores that apply, add scores together, and nally subtract Apgar
score to get morbidity score.
Apgar score
Risk A B at 5 min
1) Chronic maternal illness 13
2) Hypertension 3
3) PROM 19
4) Amniotic uid
meconium 2 or 3 6
meconium 4, fetid, poly- or 12 9 1
oligohydramnios, or bloody
5) Gestational age (weeks)
3537 56 18 2
3840 51 27 3
41 56 36 4
6) Birth weight (kg)
2,0002,499 24 46 5
2,5003,099 23 55 6
3,1003,749 22 64 7
3,7504,349 23 73 8
4,3504,849 24 82 9
4,850 26 91 10
7) Resuscitation, give
Oxygen 2
Positive pressure resuscitation 3
Tracheal intubation 5
Drugs or cardiac massage 6
8) Breathing pattern
a
Brady or tachypnea 4
Apnea 9
9) Obstetric trauma
Caput 6
Cephalhematoma 12
Bruising (anywhere) 18
Fractures 24
Total score of discriminant function AB. Total score 7 low risk;
total score 7 high risk.
a
At 6 h after delivery.
Zapata-Vazquez et al. / Archives of Medical Research 34 (2003) 214221 220
Appendix 3. Thirty nine positive morbidity patients and their respective pathologies
Score Frequency Pathology
8 1 1) Pulmonary adaptation syndrome, pathologic jaundice, impetigo
10 2 1) Hypoglycemia, 2) polycythemia
11 2 1) Sepsis, 2) purulent conjunctivitis
13 1 1) Purulent conjunctivitis
14 1 1) TTN, pathologic jaundice
16 3 1) Hypoglycemia, 2) fracture of clavicle, subconjunctival hemorrhage, 3) sepsis
17 1 1) Sepsis, pathologic jaundice
19 5 1) Hypoglycemia, apnea, died, 2) infant of diabetic mother, TTN, sepsis, 3) sepsis, 4)
purulent conjunctivitis, pathologic jaundice, 5) TTN
20 1 1) Hypoglycemia, sepsis
24 1 1) Meconium aspiration syndrome
26 4 1) RDS, 2) asphyxia, 3) anesthesia-induced depression, 4) TTN
30 1 1) TTN, sepsis, hyponatremia, intrauterine pneumonia
31 1 1) IRDS, died
38 1 1) HIE, hypoglycemia, hyponatremia
41 1 1) ELBW, died
42 1 1) Double cephalhematoma, pathologic jaundice, conjunctivitis
44 1 1) IRDS, PDA, interstitial emphysema
50 1 1) HIE, hypoglycemia
51 1 1) HIE
56 1 1) IRDS, pneumothorax, HIE, sepsis, pathologic jaundice
59 3 1) Thrombocytopenic neonatal, 2) and 3) ELBW, died
62 1 1) Intrauterine pneumonia, pathologic jaundice
72 1 1) RDS, intrauterine pneumonia, pulmonary hemorrhage
77 1 1) Meconium aspiration syndrome, HIE, seizures
89 1 1) HIE, obstetric trauma, died
Abbreviations: TTN transient tachypnea of the newborn; RDS respiratory distress syndrome; IRDS infant respiratory distress syndrome;
HIE hypoxic-sischemic encephalopathy; ELBW extremely low birth weight; PDA patent ductus arteriosus.
References
1. Hulsey TC, Hudson MB, Pittard WB. Predictors of hospital post-
discharge infant mortality: implications for high-risk infant follow-up
efforts. J Perinatol 1994;14(3):219225.
2. Richardson DK, Tarnow-Mordi WO, Escobar GJ. Neonatal risk scoring
systems. Can they predict mortality and morbidity? Clin Perinatol
1998;25(3):59611.
3. Ritter C, Bouckaert A, Van Lierde M, Theunissen I. Modeling mortality
and morbidity of newborns. Stat Med 1995;14(20):21912204.
4. Sinclair JC, Tudehope DI. Peso de nacimiento, edad gestacional y
riesgo neonatal. In: Fanaroff AA, Martin RJ, Merkatz IR, editors.
Behrman enfermedades del feto y del RN. Buenos Aires, Argentina:
Panamericana;1985. pp. 251262.
5. Klaus MH, Fanaroff AA. Asistencia del recien nacido de alto
riesgo. Buenos Aires, Argentina: Panamericana;1981. p. 406.
6. Wolf EJ, Vintzileos AM, Rosenkrantz TS, Rodis JF, Salaa CM, Pez-
zullo JG. Do survival and morbidity of very-low-birth-weight infants
vary according to the primary pregnancy complication that results in
preterm delivery? Am J Obstet Gynecol 1993;169:12331239.
7. Yancey MK, Duff P, Kubilis P, Clark P, Frentzen BH. Risk factors for
neonatal sepsis. Obstet Gynecol 1996;87:188194.
8. Berkus MD, Langer O, Samueloff A, Xenakis EM, Field NT,
Ridgway LE. Meconium-stained amniotic uid: increased risk for
adverse neonatal outcome. Obstet Gynecol 1994;84:115120.
9. Singh M, Narang A, Bhakoo ON. Predictive perinatal score in the
diagnosis of neonatal sepsis. J Trop Pediatr 1994;40:365368.
10. Perlman JM, Risser R. Can asphyxiated infants for neonatal seizures
be rapidly identied by current high-risk markers? Pediatrics 1996;
97:456462.
11. Portman RJ, Carter BS, Gaylord MS, Murphy MG, Thieme RE, Mer-
enstein GB. Predicting neonatal morbidity after perinatal asphyxia: a
scoring system. Am J Obstet Gynecol 1990;162:174182.
12. Hobel CJ, Hyvarinen MA, Okada DM, Oh W. Prenatal and intrapartum
high-risk screening. Am J Obstet Gynecol 1973;117:19.
13. Klaus MH, Fanaroff AA. Care of the high-riskneonate. Philadelphia, PA,
USA: Saunders Company;1970. p. 312.
14. Armitage P. Multiple regression and multivariate analysis. In: Armi-
tage P, editor. Statistical methods in medical research. Oxford, UK:
Blackwell Scientic Publications;1980. pp. 302347.
15. Kleinbaum DG, Kupper LL, Muller KE. Two-group analysis discrimi-
nant. In: Kleinbaum DG, editor. Applied regression analysis and other
multivariable methods. Belmont, CA, USA: Duxbury Press;1988. pp.
560594.
16. Jolicoeur P. Fishers linear discriminant function. In: Jolicoeur P, editor.
Introduction to biometry. New York: KA/PP;1999. pp. 303333.
17. Hanley JA, McNeil BJ. The meaning and use of the area under a
receiver operating characteristics (ROC) curve. Radiology 1982;
143:2936.
18. Divan MY, Harland B, Nigel H, Otter Lad PO, Western M. Fetal and
neonatal mortality in the post-term pregnancy: the impact of gestational
age and fetal growth restriction. Am J Obstet Gynecol 1998;
178(4):726731.
19. Cabals LA, Karrison T, Brown L. Factors inuencing neonatal out-
comes in the very-low-birth-weight fetus (1500 grams) with a breech
presentation. Am J Obstet Gynecol 1994;171:3542.
20. Tarnow-Mordi W, Ogston S, Wilkinson A, Reid E, Gregory J, Saud M,
Wilkie R. Predicting death from initial disease severity in very low
birth weights infants: a method for comparing the performance of
neonatal units. Br Med J 1990;300:16111614.
Discriminant Function That Predicts Neonatal Morbidity 221
21. Williams MC, OBrien WF. A comparison of birth weight and weight/
length ratio for gestation as correlates of perinatal morbidity. J Perinatol
1997;17(5):346350.
22. McCowan LM, Bruits RG, North RA, Gable G. Perinatal morbidity
in chronic hypertension. Br J Obstet Gynecol 1996;103:123129.
23. Joachim-Roy H, Flores-Nava G. Development of high risk infants
following premature membrane rupture in a neonatal intensive care
unit. Ginecol Obstet Mex 1995;63:119122.
24. Myles TD, Strasssner HT. Amniotic uid distribution in predicting
perinatal outcome in patients with ruptured membranes. Obstet Gynecol
1997;89:723728.
25. Dizon-Townson D, Kennedy KA, Dildy GA, Wu J, Egger M, Clark
SL. Amniotic uid index and perinatal morbidity. Am J Perinatol
1996;13(4):231234.
26. Richardson DK, Phibbs CS, Gray JE, McCormick MC, Wolkman-
Daniels K, Goldmann DA. Birth weight and illness severity: inde-
pendent predictors of neonatal mortality. Pediatrics 1993;91:969975.
27. Richardson DK, Gray JE, McCormick MC, Wolkman-Daniels K, Gold-
mann DA. Score for neonatal acute physiology: a physiologic severity
index for neonatal intensive care. Pediatrics 1993;91:617623.
28. Britton JR, Britton HL, Beebe SA. Early discharge of the termnewborn:
a continued dilemma. Pediatrics 1994;94:291295.
29. Edmonson MB, Stoddard JJ, Owens LM. Hospital readmission with
feeding-related problems after early postpartum discharge of normal
newborns. JAMA 1997;278:299303.
30. Parrish KM, Holt VL, Connell FA, Williams B, Logerfo JP. Variations
in the accuracy of obstetric procedures and diagnoses on birth records in
Washington state, 1989. Am J Epidemiol 1993;138:119127.
31. Kotelchuck M. An evaluation of the Kessner adequacy of prenatal care
index and a proposed adequacy of prenatal care utilization index. Am
J Public Health 1994;84:14141420.
32. The International Neonatal Network. The CRIB (clinical risk index for
babies) score: a tool for assessing initial neonatal risk and comparing
performance of neonatal intensive care units. Lancet 1993;342:193198.
33. Courcy-Wheeler RH, Wolfe CD, Fitzgerald A, Spencer M, Good-
man JD, Gamsu HR. Use of the CRIB (clinical risk index for babies)
score in prediction of neonatal mortality and morbidity. Arch Dis Child
Fetal Neonatal 1995;73:3236.
34. Pollack MM, Patel KM, Ruttimann UE. PRIMSIII: an updated pediatric
risk of mortality. Crit Care Med 1996;24:743752.
35. Knaus WA, Draper EA, Wagner DP, Zimmerman JE. APACHE II: a
severity of disease classication system. Crit Care Med 1985;13:818829.
36. Gordis L. Assessing the validity and reliability of diagnostic and screen-
ing test. In: Gordis L, editor. Epidemiology Philadelphia, PA, USA:
WB Saunders Co.;1996. pp. 5875.
37. Riegelman RK, Hirsch RP. Como estudiar un estudio y probar una
prueba: lectura cr tica de la literatura medica. Bol Sanit Panam
1991;111(6):548555.
38. Bates AS, Margolis PA, Evans AT. Verication bias in pediatric studies
evaluating diagnostic test. J Pediatr 1993;122(4):585590.
39. American Academy of Pediatrics. Committee on Fetus and Newborn.
Criteria for early infant discharge and follow-up evaluation. Pediat-
rics 1980;65:651.
40. Liu LL, Clemens CJ, Shay DK, Davis RL, Novack AH. The safety of
newborn early discharge. JAMA 1997;278:293298.