Contents - Previous - Next
WHO  has conducted numerous clinical studies and mortality surveys using an algorithm for tetanus diagnosis based on the questions in Table 2.
The validity of verbal autopsy diagnoses of neonatal tetanus is difficult to assess directly. In larger clinical series, cessation of suckling is reported in 69% to 100% of clinical cases, rigidity or spasms in 97% to 100%, and trismus/risus sardonicus in 71% to 100% of cases. Fever or umbilical sepsis is reported in 20% to 50%, . One study in Cebu found a sensitivity of 94% to 100% using modification of the WHO algorithm, but the specificity could not be adequately assessed . In a Bangladesh study, the WHO algorithm was compared to all possible cases of tetanus identified by open ended interviews with mothers of deceased children, and the WHO algorithm had a sensitivity of 52% and a specificity of 100%. The low sensitivity was due to the fact that many women do not breast feed for some days after birth, so the question on cessation of feeding had little predictive value in this context . Also, the absence of clinically confirmed diagnoses marked this Bangladesh study difficult to interpret.
Table 2. Algorithm for Neonatal Tetanus
1. Did the infant die between the 3rd and 30th day of life?
There is evidence of epidemiologic plausibility for the verbal autopsy diagnosis in the age distribution of putative tetanus deaths as compared to other perinatal deaths, and in urban/rural or seasonal differentials of estimated disease incidence. Also, use of algorithm diagnosis in evaluation of preventive programmes of maternal tetanus toxoid vaccination or traditional birth attendants (TBA) training in cord care has shown a substantial decline in diagnosed neonatal tetanus deaths as compared to control populations with no interventions [9,10].
Deaths during the perinatal period are often not detected because of failure of maternal recall, or because survey interviews ask questions only about live births, and thus omit stillbirths. Also, some early neonatal deaths during the first week of life are probably misclassified as stillbirths . Nevertheless, a number of investigations suggest that perinatal mortality is a major and relatively poorly understood public health problem in the developing world. It is, however difficult to diagnose causes of death during the perinatal period, particularly if delivery occurred without medical supervision.
There are no well established questions that have been used to diagnose perinatal causes of death, and the suggested items in Table 3 are derived from a number of reports.
Table 3. Perinatal Deaths
1. Did the baby show any sign of life after birth (i.e., breathing, crying)?
2. Did the baby die within one week of birth?
3. How big was the infant at birth?
Normal size ,______
4. What was the duration of the pregnancy? _______
5. Was the delivery prolonged (> 12 hours)?
6. Were there bruises or marks of trauma on the baby's body?
7. Was this a multiple pregnancy?
8. Did the baby fail to breathe, cry or suckle normally after birth?
9. Did the baby have spasms or convulsions within the first three days of life?
10. Was the baby floppy or unconscious during the first 3 days of life?
11. Did the baby have any abnormalities of the head, body or limbs?
Affirmative responses to questions 1 and 2 establish a perinatal death, questions 3 and 4 suggest prematurity/low birth weight, questions 5 to 7 birth trauma/asphyxia, and questions 8 to 10 differentiate perinatal problems from neonatal tetanus. Question 11 attempts to identify congenital effects.
The WHO EPI programme  and other investigations have used verbal autopsy methods to diagnose deaths associated with measles. The most common questions are listed in Table 4.
Measles is often recognized by the mother and there are frequently local terms for the disease that can be useful in diagnosis. But such local terminology may be applied to other illnesses that present rashes and, by itself, may be too non-specific for a diagnosis. The occurrence of a measles epidemic or of other cases in the household also helps to establish measles as a probable cause. A history of measles vaccination makes the diagnosis less likely, but does not exclude the disease because vaccine failures are known to occur.
Table 4. Diagnostic Algorithm for Measles
1. Death of a child after the 5th month of life
Several studies have shown that verbal autopsy diagnosis of measles is highly reliable. A study of 48 measles-associated deaths in Cebu compared maternal interviews with physician diagnoses, and the verbal autopsy diagnosis was found to have 98% sensitivity and 90% specificity . A similar study of children who survived acute measles found a sensitivity and specificity of 94% . In a Kenyan study, serological tests for measles antibodies were compared with maternal interview diagnoses for 556 children; the sensitivity was 96%, but the specificity was 54% for all "possible, probable and definite clinical cases." If only "definite" cases are considered (i.e., those meeting all criteria in the above algorithm), the specificity was 85% . However, a sero-epidemiologic study of 600 unvaccinated children in Mozambique found that a maternal history gave 50% sensitivity and 83% specificity . Several studies have shown declines in verbal autopsy-diagnosed measles death following immunization programmes, which adds epidemiologic plausibility .
Most studies of diarrhoeal morbidity or mortality define an episode of diarrhoea as a history of three to four loose liquid stools per day, and a history of dysentery is defined as frequent liquid stool containing blood. Open-ended questions on the presence of "diarrhoea" may be unreliable, because of the subjective nature of maternal assessment and because of cultural variation in the use of the local terms [17,18]. Since diarrhoeal deaths are generally associated with dehydration, interviews usually ask questions about such signs as thirst, sunken eyes, dry mouth etc. [19,20]. A suggested list of questions is given in Table 5.
Table 5. Diagnostic Algorithm for Diarrhoea/Dysentery
1. History of 3-4 or more loose or liquid stools per day?
A study in Cebu compared verbal autopsy diagnoses of diarrhoea associated deaths with confirmed clinical diagnoses in 43 children who died as a result of diarrhoea alone, and 92 children who died of diarrhoea plus other associated illnesses . Among the children in whom diarrhoea was the sole cause of death, a history of loose or liquid stools had a sensitivity of 84% and a specificity of 79%. Further questions indicating severe diarrhoea (³ 6 stools per day, thirst, or sunken eyes) decreased the sensitivity to 68%, but increased the specificity to 92%. In the 92 children who died of diarrhoea and another concomitant disease, the history of loose/liquid stools had a sensitivity of 78% and a specificity of 79%; with questions on more severe diarrhoea, the sensitivity was reduced to 57% but the specificity increased to 92%. A related study of children who survived a diarrhoeal episode showed high sensitivity (95%) and specificity (80%) for interview-based diagnoses, but the diagnosis of dehydration was less satisfactory except in the most extreme cases . It is difficult to assess the degree of fluid loss without weighing the child before and after rehydration, and even physicians underestimate the severity of dehydration when using clinical judgment rather than objective weight change .
A study in Bangladesh compared maternal histories of diarrhoea with findings of stool cultures in living children, and found an 80% agreement . However, in a Kenyan study maternal reports of "diarrhoea" (without specifying the number or consistency of stools) were compared to visual inspection of stool specimens, and it was found that mothers over-reported 15% to 40% of episodes . This highlights the need for detailed information of stool characteristics and frequency rather than subjective maternal assessment. A Gambian investigation of children with marasmus and chronic diarrhoea showed that maternal interviews on admission correctly identified 75% of cases, and one month after discharge, maternal interviews agreed with the clinical diagnosis in 91% of cases . However, the post discharge interview could be affected by communication with the physician during hospitalization.
Field studies using maternal interviews have shown age, sex, and seasonal patterns of diarrhoea consistent with those observed in clinical studies [17, 21, 23]. Also, ORT programmes have demonstrated declines in diarrhoeal mortality consistent with observations from field trials [24-26]. This suggests epidemiologic plausibility for interview based diagnoses.
In summary, verbal autopsy diagnoses of diarrhoea appear to be reliable if adequate information is obtained on stool frequency and consistency. However, maternal recognition or recall of the severity of dehydration may be problematic.
The diagnosis of ALRI is difficult because the symptoms and signs of respiratory infection vary with age and with severity of illness. Ideally, definitive clinical diagnosis should be based on X-ray findings, sputum culture for causative organisms, or specialized tests (such as pulse oximetry) that measure blood oxygen levels. Without such investigations, there is uncertainty in physician diagnoses due to interobserver variation in the detection of respiratory signs during auscultation of the chest [27-29].
The algorithm in Table 6 is based in part on clinical judgement, and has not been fully evaluated.
A validation study in Cebu examined 100 ALRI associated deaths and found that the diagnosis was problematic . The presence of cough and dyspnoea without qualifying the duration of symptoms had high sensitivity (86%) but low specificity (47%), whereas cough lasting four or more days and dyspnoea for one or more days had a sensitivity of 59% and a specificity of 77%. The main difficulty is that all children who die show signs of terminal respiratory distress that can only be differentiated from ALRI by questions specifying duration of symptoms for more than one day. Also, many severely ill children with non-ALRI respiratory distress have a cough, which may be difficult to distinguish from ALRI without specifying duration (at least 3 to 4 days). We know of no other study to validate verbal autopsy diagnoses of ALRI. However, field trials of pneumococcal vaccine and studies of antibiotic treatment that used verbal autopsy diagnoses have shown declines in ALRI- associated mortality, which suggests that the diagnosis is sufficiently accurate to detect the impact of interventions [30,31].
Table 6. Diagnostic Algorithm for ALRI
1. Cough (³ 3-4 days)
2. Difficulty breathing (dyspnoea) as indicated by chest indrawing, nasal flaring, grunting or wheezing, cyanosis (blue color of lips), lasting ³ 1 day
3. Rapid breathing
There have been several investigations of diagnostic accuracy in children with ALRI who did not die of the disease. Studies that compare physicians' findings in sick children with x-ray confirmation of ALRI or pulse oximetry measurements suggest that the clinical diagnosis itself may be unreliable. For example, in studies of young infants, a rapid respiratory rate (>60 per minute) had sensitivities ranging from 40% to 61%, and specificities ranging from 55% to 61%. Similarly, chest retractions had a sensitivity range of 66% to 94% and a specificity range of 28% to 54% . In older children rapid respiration (>50 per minute) had higher sensitivity (66% to 75%) and specificity (67% to 98%), but these investigations were conducted in selected groups of sick children with suspected pneumonia . A field study in Tari, Papua New Guinea, showed 77% agreement between clinical diagnosis and X-ray findings [Dr. Lehman personal communication 1990].
Maternal interview based diagnoses of ill children with ALRI have been evaluated in a number of investigations. A British study interviewed mothers whose children had physician diagnosed ALRI during infancy, and found that only 17% of mothers remembered the illness when questioned. at the child's first birthday . In the Gambia, clinical diagnoses were compared with maternal interviews at the time of hospitalization and were found to agree in 84% of cases . Also in Cebu, interviews with mothers of children with ALRI who had been discharged from hospital, were compared to physicians' diagnoses . A combination of cough, fever, and dyspnoea gave a sensitivity of 68% and a specificity of around 84% for the maternal interview diagnoses. A study in Burkino Faso found that ALRI diagnoses by field worker interviews underestimated ALRI incidence as compared to physician diagnoses .
In summary, efforts to validate ALRI diagnoses based on verbal autopsy interview, physicians' clinical findings or interviews with mothers of surviving children have shown that there is substantial misclassification of children with ALRI. This raises concerns about the reliability of studies of ALRI mortality and morbidity.
The definitive diagnosis of pertussis is based on culture of the organism or serological tests, but a presumptive clinical diagnosis can be made on a history of paroxysmal coughing persisting for two or more weeks, with choking and/or vomiting after the paroxysm. There may also be a characteristic "whooping" sound during inspiration, and children frequently have conjunctival haemorrhages. The following algorithm (Table 7) is based on WHO studies  and studies in Kenya [34-35].
Table 7. Algorithm for Pertussis
1. History of severe paroxysms of coughing ³ 14 days
There has been no direct validation of the above questions used for verbal autopsy. Studies of sick children have validated these criteria for the diagnosis of morbidity by comparing probable or definite diagnosis of pertussis based on laboratory findings with diagnoses based on maternal history alone. Investigators in Kenya  found confirmatory laboratory evidence for pertussis in over 80% of suspected cases during an epidemic, but in only 40% of sporadic cases during non-epidemic periods. A clinical study in the U.S.A. found that a history of prolonged cough (214 days) had a sensitivity of 98% and a specificity of 63%; additional criteria for paroxysmal coughing decreased the sensitivity to 52% but increased the specificity to 83% . Several vaccine trials using interview-based diagnoses have shown declines in pertussis morbidity and mortality [9,34,35]. It is therefore likely that the algorithm for pertussis is satisfactory in field settings.
The clinical diagnosis of malaria can be difficult, and a definitive diagnosis depends on demonstration of a high density of parasites in the blood. In Gambian studies, a presumptive diagnosis was made on the basis of a history of high fever in previously well children, without significant other concomitant respiratory or gastrointestinal illness . If malaria transmission is seasonal or epidemic, then a febrile illness during periods of peak transmission is more likely to be due to malaria. It is difficult to assess the validity of this presumptive diagnosis, in part because many children in endemic malarious areas have parasitaemia without fever, and children who have received antimalarial drugs have no parasitaemia. Morbidity surveys in the Gambia suggest that around 65% of children with fever have malaria parasites in their blood, but 30% of a febrile children also have parasitaemia .
Similarly, a study of illness in Congolese school children showed that 46% of children with fever had evidence of malaria, and among 153 episodes of illness, malaria was suspected in 43% of cases, but the diagnosis could only be confirmed in 20% . Physician diagnoses are also problematic; a study of 104,281 patients in Zimbabwe showed that the percentage of clinically diagnosed cases of malaria confirmed by laboratory tests varied from under 5% in the season of low transmission to over 30% in the season of peak transmission . Thus, fever as a criterion for malaria diagnosis is somewhat insensitive and nonspecific. Nevertheless, one investigation of 16 seriously ill Gambian children with clinically diagnosed malaria found that open-ended interviews with mothers gave a presumptive diagnosis of malaria in 75% of cases on admission to the hospital and in 90% of cases one month after discharge  . In summary, there is insufficient information to assess the accuracy of the diagnosis of malaria deaths by verbal autopsy methods, and this is a priority area for research.
Table 8. Algorithm for Chronic Conditions
1. Weight loss/slow growth (malnutrition, TB, AIDS)
Criteria for verbal autopsy diagnosis of meningitis have not been well established, but studies in Senegal and Nigeria used an algorithm based on fever, headache, neck extension (often with flexed arms and legs), and swollen fontanelle in infants [5,40]. Additional signs included convulsions and photophobia. Since most deaths occur during epidemics, the presence of other cases provides supporting evidence. No validation studies for the interview diagnosis of meningitis are available.
Many children suffer from chronic ill health as a consequence of nutritional deficiencies, recur-rent acute infections, and chronic infections with diseases such as tuberculosis. Pediatric AIDS is also a growing public health problem, particularly in Africa. It is, however, difficult to distinguish these chronic conditions from one another and hard to determine their contribution to death using verbal autopsy methods.
There is no satisfactory algorithm for the verbal autopsy diagnosis of chronic malnutrition, TB, or AIDS. The questions listed in Table 8 draw upon several studies but have not been adequately evaluated. For each question, the possible associated diseases are given in parentheses.
It is relatively easy to obtain information on deaths due to injury using verbal autopsy methods, by asking descriptive questions on how the injury occurred (external cause, e.g., car accident) and on the nature and location of the injury (e.g., fractured skull). However, there may be underreporting of causes such as homicide (particularly infanticide) and suicide, and there may be imprecise descriptions of the nature and location of injury. Confirmation of deaths due to injury may also be obtained from medical or legal records .
WHO  has classified the external causes of injury into nine categories as shown in Table 9.
The use of verbal autopsy methods and this classification of causes of injury has not been validated.
Table 9. External Causes of Injury
1. Bites or stings of venomous animals
In the absence of reliable information on causes of death in developing countries, researchers and policy makers will depend upon information derived from secondary sources such as interviews with relatives of ill or deceased individuals. Such verbal autopsy diagnoses will always be subject to error because few diseases present as unmistakably distinct syndromes with symptoms and signs that can be readily perceived and recalled by the relatives. Moreover, the absence of clinical information will, of necessity, result in less precise presumptive diagnoses rather than definitive clinical diagnoses. Nevertheless, verbal autopsies can provide data of value to public health and it is likely they will be more widely used in surveillance and survey studies. There is a need for greater standardization of interview methods and wording of specific questions to provide more consistency among investigations. Also, there is a need for anthropologic studies to help define culturally specific constructs and terminology. Finally, more research is needed to validate interview diagnoses against clinical diagnoses in a variety of cultural settings.
Contents - Previous - Next