Validity of International Classification of Diseases (ICD) coding for dengue infections in hospital discharge records in Malaysia

Dengue has become a global public health concern. Epidemiologic measures of the burden of dengue such as its incidence and prevalence are important for policy-making and monitoring the progress of disease control. WHO reported the global incidence of dengue has increased 30-fold in the past 50 years and estimated some 50 to 100 million new infections occurred annually [1], causing about 20,000 deaths [2]. These estimates are largely based on dengue notification data reported to national surveillance systems, which are widely used as a proxy measure of dengue incidence [3‐7].

However, for dengue where the majority of infected individuals are asymptomatic and may suffer no or little adverse health consequences, estimates of incidence and prevalence are measures of disease frequency rather than measures of the disease burden. Case fatality rate of dengue remains low with an average of 2.4 deaths over 100,000 notified case in Malaysia [8], but symptomatic dengue infection is significantly associated with considerable morbidity. Hospitalization due to an acute illness like symptomatic dengue infection is an important measure of morbidity. Dengue hospitalization is a critical driver of economic cost of the illness to society, especially in a country endemic of dengue. Apart from the medical costs, hospitalization also leads to loss in economic productivity arising of sick workers taking sick leaves or parents taking time off to care for their sick children. Nonetheless, health services utilization associated with dengue illness, and hospitalization in particular, remains poorly characterized, though there have been many hospital based, mostly single-center, studies on dengue [9‐17]. Few studies have make use of administrative data which are available from hospital discharge and health insurance claims databases [18] in part because the validity of discharge diagnoses codes is uncertain. We therefore undertook this study to evaluate the accuracy of discharge diagnoses coded according to the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes for dengue. Validation of these codes is necessary to promote the wider use of hospital discharge and insurance claims data for health services research on dengue.

This validation study extracted data from the available hospital discharge records and retrospective review of medical records. The Medical and Research Ethics Committee approved the study (NMRR-15-452-25,624).

Table 1 shows the characteristics of the patients with and without ICD-10-CM codes for dengue whom were included in this study. Their mean age were 37 and 28 years respectively. About half the patients were diagnosed in year 2010, and the other half in 2013. Bleeding, visceral signs, low platelet count, raised blood hematocrit were more commonly present among patients with ICD-10-CM codes for dengue. Organ impairment however were found equally between the two groups. Table 2 summarizes the number of cases with positive result for NS1 and/or IgM among patients with and without ICD-10-CM codes for dengue. 87% of those with the dengue codes had positive laboratory results (NS1 or IgM) for dengue while only 8% of those without dengue codes had positive results.

Table 3 shows the diagnostic performance of the ICD-10-CM codes for dengue. The sensitivity was 94% (188/ 200) and specificity 83% (132/160), giving an overall accuracy of 88.9% (320/ 360). The positive Likelihood ratio (LR) was 4.7 and negative LR 0.075. This provides adequate evidence for the diagnostic validity of ICD dengue codes. Its PPV was 87% (188/ 216) and NPV 92% (132/ 144). The diagnostic performance of dengue codes did not change between 2010 and 2013 (Table 3). Neither did it differ between the sexes. However, the specificity of ICD-10-CM codes for dengue decreased substantially when the patients presented with bleeding manifestations or low platelet count.

Administrative data such as hospital discharge and health insurance claims databases have infrequently been used for health services research on dengue despite their considerable strengths which included nationally representative sample to allow generalizability, larger sample size and low cost. The main disadvantage of administrative data is uncertain validity of the ICD codes used to identify patients with dengue. This study was undertaken to address this weakness and is the first to investigate the validity of ICD-10-CM codes for dengue. Overall we demonstrated adequate diagnostic evidence to support the use of ICD-10-CM codes in identifying hospitalized patients with positive NS1 or IgM test for dengue.

The sensitivity of ICD-10-CM codes for dengue is high (94%), but its specificity is more modest at 83% (17% false positive ratio). The lower specificity could have two explanations. Firstly, despite advances in dengue diagnostic tests [19], the diagnosis of dengue in current practice is still clinically based rather than entirely dictated by positive NS1 or IgM lab results [20]. This is further supported by our findings of lower specificities when patients manifested with bleeding or low platelet count, both commonly associated with dengue. In other words, patients are more likely to be clinically diagnosed of having dengue even when their NS1 or IgM tests were negative. Secondly, our records selection process also contributed to this modest specificity result. We have deliberately restricted selection of records without ICD codes for dengue to B34.9, A83, R50, A92 to A99 codes because they are most likely a priori to be misclassified as false-positive, thus lowering the specificity estimates. If we would have randomly selected all records without ICD codes for dengue as controls, we would then find nearly all to be true negative given the low prevalence (< 3%) of dengue among all hospitalized subjects, thus inflating the estimate of specificity.

Our results have several implications. For the use of administrative data in descriptive studies to estimate the incidence or prevalence of patients hospitalized for dengue illness, the high sensitivity of A90–91 codes ensures that most of the dengue patients will be identified. However, its moderate specificity (83%) will inflate the estimate of the true frequency of dengue in the population, with high number of false positives. Even moderate loss in specificity can lead to substantial over-estimation especially when the incidence or prevalence of the condition under investigation is low. Our finding that the validity of the ICD-10-CM codes has been stable over time is reassuring for studies on the secular trend in dengue incidence or prevalence.

For analytic studies, the ICD-10-CM codes can be used to sample cases of dengue and non-dengue controls to investigate outcome of subjects’ exposure to dengue. Assuming non-differential misclassification (ie, classification by ICD-10-CM codes is independent of the outcome of interest) and assuming having dengue increases risk of the outcome, the modest specificity (83%) of A90–91 codes will cause cases to have more false positive subjects among them to dilute the risk estimate of the case group, while the high sensitivity (94%) will minimize the number of false negative subjects in the control group, thus lessening its risk inflation. The end result will be a biased estimate of the relative risk towards null.

We conclude that the hospital discharge database maintained by the Ministry of Health (MOH) Malaysia is potentially a useful data resource to support health services research for dengue in Malaysia. The adequate diagnostic performance of the ICD-10-CM coding in the national hospital discharge database in identifying patients with dengue makes it particularly acceptable for use in research on dengue.