Piloting an event-based surveillance model in private hospitals for early detection of disease clusters, Kerala, India

Study setting

Kerala, the southernmost State of India, has 14 administrative districts and a population of 35 million. With over 1.8 million international passengers arriving annually through the four international airports, the State faces a heightened risk of novel pathogens entering. Kerala has experienced frequent outbreaks of infectious diseases, including West Nile Virus, Kyasanur Forest Disease, Nipah, Zika, Covid-19 and MPox^13-18.

We piloted the EBS system in Kasaragod, a northern district of Kerala with a population of 1.5 million, between May and December 2023. The district is characterised by local communities living close to the Western Ghats, a densely forested mountain range, and a significant proportion of young and middle-aged people working abroad. We selected six private hospitals in the district with a minimum of 60 inpatients, five intensive care beds and the highest inpatient admissions in the preceding year.

The event-based surveillance (EBS) pathway

Our hospital-based EBS focused on monitoring four syndromes: severe acute respiratory illness (SARI), acute encephalitis syndrome (AES), acute febrile illness with haemorrhage (AFIH) and acute febrile illness with rash (AFIR). These syndromes cover the typical manifestations of many emerging infectious diseases. Standard case definitions to define the syndromes^19-21 were used (Box 1). Following two rounds of detailed stakeholder discussions involving physicians, pulmonologists, intensivists, and public health experts in the State, we finalized seven signals for monitoring through the EBS (Box 2).

We routinely collected data from all in-patients with acute febrile illness (AFI) with a documented temperature of ≥38^°C or a history of fever within the last 10 days from the internal medicine, pulmonary medicine, obstetrics and gynaecology, paediatrics, and intensive care units of the selected hospitals. Data on demographics, initial clinical features, and a provisional diagnosis of the admitted AFI patients were extracted daily in a standardised case reporting format from the clinical records by trained clinical nurses using an electronic tablet. Once entered into a database, we used an algorithm to analyse the database to identify signals automatically. If any of the signals (Box 1) were detected, the information was emailed to the District Surveillance Unit, study hospital, and principal investigator and displayed on the dashboard automatically by the software. The epidemiologist at the DSU of IDSP verified the signals by reviewing the EBS data and gathering more information from the field through field workers (health inspector and public health nurse) of the concerned wards. A verified signal was called as an event. Once flagged as an event, the DSU did the risk assessment to classify the events as cluster/outbreak and initiated response measures as per the standard operating procedures of the IDSP²². The EBS pathway is illustrated in figure 1.

Close

Fig. 1.

The event-based surveillance (EBS pathway). Source: Open-source icons downloaded from Freepik ( https://www.freepik.com ).

Export to PPT

A laboratory testing algorithm was also developed for testing clinical samples when an event was detected through EBS. The laboratory testing algorithm was adapted from the guidelines of the Viral Research and Diagnostic Laboratory Network (VRDLN) of India²³. Supplementary table I shows the various pathogens tested when an event was detected.

The EBS platform

A web-based signal detection platform was developed. The platform had two portals: a data entry portal and an admin portal. The data entry portal was designed for hospitals to upload patient information using unique login credentials in a standardised case reporting format (CRF). The admin portal was developed for the monitoring by the DSU staff and the principal investigator (PI). The feature-oriented platform could perform fully scripted automated signal detection, signal visualisation, event updating, and email-based alerting of the hospitals, DSU, and PI.

Enrolment of patients

The EBS was piloted during May 23, 2023, through December 23, 2023. During the pilot, we collected data from 3294 (73%) of the 4512 patients hospitalised with febrile illness in the six hospitals. A total of 894 (19.8%) patients did not consent to participate in the study, 212 (4.7%) were missed by the data collectors during the initial phase, and 112 (2.5%) were referred to another centre before data collection. The patients enrolled in the EBS were from 40 of 41 local self-government units (LSGU) in the district. The median (IQR) age was 20 (5-54) yr, with an almost equal sex distribution. The most reported comorbid conditions among patients over 30 yr were hypertension (n=463, 34%) and diabetes mellitus (n=350, 26%; Table I).

Description of signals

We identified 88 signals during the implementation of EBS. Figure 2 shows the timeline of signals and events. At least one signal was detected every week except for epi-weeks 21, 22, 26, and 51. Epi-week 33 detected the largest number of signals (n=12, 14%). Most signals were due to SARI (n=67, 76%), followed by AES (n=9, 10%) and AFIR (n=8, 9%). In addition, three healthcare worker signals were detected on the 23^rd, 24^th, and 28^th epi-week, and one signal due to a death on epi-week 24 (Table II and Fig. 2). No AFIH signals were detected during the study period. Signals were detected in 28 (68%) LSGUs (Fig. 3). About one-fourth of the signals (n=20, 23%) were from three urban LSGUs in the district. Twenty-eight (32%) signals were detected from the four rural LSGUs, which were adjacent to these urban areas.

Close

Fig. 2.

Timeline of signal and events detected through the EBS May-December 2023.

Export to PPT

Close

Fig. 3.

Residence of patients who triggered EBS signals, by local self-government unit, and signal triggered. Source: Map created with QGIS using open-source shape file of the district available with the district administration.

Export to PPT

Event identification and investigation of outbreaks/clusters

All 88 signals were investigated by the district surveillance unit: of them, 10 signals flagged as ‘events’ following verification, and the rest were discarded. The reasons for discarding the signals are provided in supplementary table II. Nine of the events were classified as outbreaks based on risk assessment (Table III). Among the 10 events identified, eight were SARI, and one each was an AFIR and AES event. Three of the eight SARI events were due to influenza A(H1N1), and one was due to influenza A(H3N2). In two events, mixed infections (influenza A(H1N1) and influenza A(H3N2), as well as influenza A(H1N1) and SARS-CoV2, were identified. No pathogen could be isolated in two SARI events. The AFIR event was due to dengue. No pathogens could be detected for the AES event (Table III). The investigation of these events resulted in declaring nine events as outbreaks and identified 305 case-patients during field investigation. In one of the SARI events classified as an outbreak, one death was reported with a case fatality ratio of 4.5 per cent (1/22) (Table III).

During the pilot period, EBS detected five events (4 SARI and 1AFIR) that were not identified by the routine IBS. Additionally, five events (1 dengue, 1 AES, and 3 H1N1 events) were detected through both IBS and EBS.

Performance indicators of EBS

The positive predictive value (PPV)for detecting events was 11.4 per cent (10/88), and the PPV for events confirmed as outbreaks was 90 per cent (9/10). The median time interval between identifying events in EBS and confirming outbreaks in IBS was 5 (2-6) days.

Challenges in implementing the EBS

One of the major challenges we faced was onboarding private-sector hospitals. The hospitals were initially apprehensive, particularly about the perceived increased burden for staff. However, they appreciated the simplicity of the portal, which required minimal time and resource commitment. The key factors facilitating their participation in the pilot included a user-friendly data entry platform, access to advanced laboratory testing facilities for their patients in case of signals/events, and the added utility of the EBS to public health surveillance. Because our EBS was fully automated, the data entry quality was crucial for project success. To ensure data quality, we trained and utilised clinical nurses from the enrolled hospitals as data collectors with compensation for their time. This ensured the timely collection of precise data from the patient’s clinical records. Another challenge we faced during our pilot was limited training for field-level workers in outbreak response. To address this, we conducted training sessions and provided them with standard operating procedures (SOPs) to respond to outbreaks. There was also a gap in obtaining and testing samples from outbreaks detected. As part of the project, we followed the sample testing algorithms of the VRDL and utilised their logistics for sample collection, storage, transport, and testing of various viral pathogens. The core component of our project was its linkage with the district surveillance unit of IDSP. Achieving this required significant efforts and multiple deliberations with the State and district health authorities to obtain the necessary permissions. This integration was vital for ensuring that the system complemented existing surveillance mechanisms, and facilitated prompt verification of the signals and initiating public health responses for the identified events.

Our study has certain limitations. First, about 20 per cent of eligible patients refused consenting to participate in the study. This may be because the clinical nurses could not clearly explain the study to the patients, particularly in the early data-collection phase. This was addressed by training the nurses about obtaining written informed consent. As our study focused on identifying clustering of cases by time and location, we could have detected more signals if the non-response in the study was lower. Second, we selected hospitals for the pilot based on the patient load and functional intensive care units (ICUs). We assumed that patients with severe manifestations, including those with SARI, AES, and AFIH syndromes, were more likely to get admitted to these secondary/tertiary care hospitals. However, the selected hospitals might not represent the entire private sector in the district. Another limitation of the pilot was access to metagenomic next-generation sequencing (NGS) for events where the aetiology could not be identified based on serological and molecular tests. Incorporating NGS capabilities in future iterations of the EBS system could greatly enhance the ability to identify novel pathogens, facilitating the early detection of emerging infectious disease threats. Since we did not have information from the IBS portal on the total number of outbreaks in the catchment area of the study hospitals, we were unable to calculate sensitivity. However, future evaluations of full-scale EBS implementation may consider collecting data about additional attributes such as sensitivity and usefulness³¹.

Despite these limitations, the findings of our pilot indicated that establishing EBS at private-sector hospitals in Kerala complimented the IBS by early detecting outbreaks. Given its added value in strengthening disease surveillance, the IDSP may consider implementing this model in other districts, especially those at higher risk of zoonotic spillover.