The ER Is Packed, But 40% of Patients Shouldn't Be There: How AI Pre-Screening Reduces Emergency Department Overcrowding

The ER Is Full — But Not All of It Should Be

It is 9 PM on a Tuesday. In the emergency department of a large public hospital in Jakarta, every bed is occupied. Nurses move between stretchers lining the corridor. A security officer manages the queue at the entrance gate. Triage staff work through a backlog of patients, some of whom arrived hours ago with complaints that will turn out to be mild fever, runny nose, and sore throat — conditions that could have been managed at a Puskesmas or general practice clinic, if only one had been reachable at this hour.

This scene repeats every night, across hundreds of Indonesian hospitals. Emergency department (IGD) overcrowding is not a headline — it is a structural reality. And buried inside that overcrowding is a pattern that healthcare administrators know well but struggle to solve: a large share of patients presenting to the IGD are not, by clinical definition, emergencies. Internationally, research puts the proportion of non-urgent ED presentations at a median of 32%, with values in certain healthcare systems reaching as high as 82%.^[1] In the United States, roughly 30–50% of ED visits are classified as non-urgent, costing an estimated $32 billion in unnecessary healthcare spending annually.^[2]

Indonesia's IGD system faces a version of this problem that is compounded by specific structural factors: limited Puskesmas operating hours, low public trust in primary care, the dominance of walk-in culture, and the widespread perception that a hospital's emergency department is the fastest route to any kind of medical attention. One South Kalimantan hospital study found that emergency visits surged from 25.17% of total visits in 2021 to 40.86% by 2023 — a trajectory that signals a system under mounting pressure.^[3] The consequence for patients who genuinely need emergency care: longer waits, diluted staff attention, and in the most serious cases, delayed treatment that correlates with preventable adverse outcomes. Research has directly linked peak IGD overcrowding to a higher incidence of unexpected cardiac arrest.^[4]

Why Patients Go to the IGD When They Don't Have To

To fix the problem, you must first understand it. Patients who present to emergency departments with non-urgent conditions are not acting irrationally — they are responding to an information gap. They do not know how serious their symptom is. They do not know which specialist to see. They do not know if the clinic is open. They cannot get a same-day appointment. They are anxious, and anxiety amplifies perceived urgency.

In the Indonesian context, several additional dynamics accelerate this pattern:

• After-hours care deserts: Most Puskesmas and general practice clinics operate from roughly 7 AM to 2 PM. After that window closes, the IGD is effectively the only accessible care option for millions of Indonesians — regardless of the severity of their complaint.
• Primary care distrust: A significant portion of the Indonesian population bypasses Puskesmas not because they are unavailable, but because patients perceive hospital care as higher quality, faster, or more definitive. This perception drives direct-to-IGD behavior even for conditions like gastritis, mild urinary tract infections, and skin complaints.
• BPJS complexity: Many patients are unsure whether their BPJS coverage requires a referral from a primary care facility, or whether visiting the IGD directly is covered. Facing uncertainty, they choose the one door they know is always open: the emergency department.
• Symptom anxiety: A patient who woke up with chest tightness, a child with a fever of 39°C, or an elderly person who felt dizzy standing up — all of these individuals may reasonably fear the worst. Without access to guidance on whether their symptom warrants emergency care, the safest-feeling choice is to go to the IGD immediately.
• No pre-visit guidance channel: Unlike many developed healthcare systems where nurse hotlines or digital triage tools exist, most Indonesian patients have no structured way to ask “should I go to the ER?” before they show up.

“Emergency department overcrowding is not simply a hospital problem — it is a failure of the information infrastructure that should guide patients to the right level of care before they ever reach the hospital door.” — International Journal of Emergency Medicine, 2026^[5]

The Clinical and Operational Cost of Misallocated Emergency Care

When a patient who needs a two-day course of antibiotics occupies an IGD bed for three hours, the cost is not just financial — it is measured in delayed care for every patient who arrives after them. Overcrowding in emergency departments has been associated with increased patient mortality, higher rates of patients leaving without being seen, increased medication errors, and accelerated staff burnout.^[4] For Indonesian hospitals already operating under constrained human resources, absorbing a surge of non-urgent cases during peak evening hours creates a compounding crisis.

The financial dimension is equally significant. Non-urgent IGD visits that do not meet BPJS Kesehatan's emergency criteria are classified as “false emergencies” (gawat darurat semu) and may not be fully covered — leaving patients with unexpected out-of-pocket bills and hospitals with collection challenges. At the same time, hospitals absorb the overhead of triaging, treating, and documenting these visits while losing bed capacity that should be reserved for genuinely acute cases.

For nursing and medical staff, the human cost is real. IGD nurses in Indonesia are frequently required to manage simultaneous triage, acute resuscitation, and administrative intake — a workload that becomes untenable when the department is flooded with low-acuity cases. Staff retention in emergency medicine is an ongoing challenge globally, and overcrowding accelerates the burnout trajectory.

Routing Before the Door: The Case for AI Pre-Screening via WhatsApp

The most effective intervention for IGD overcrowding is not inside the emergency department — it is before the patient leaves home. If a patient can receive structured, safe guidance on whether their symptom warrants emergency care within two minutes of messaging a trusted channel, the downstream effect on IGD volume is direct and measurable. This is the core premise of AI pre-screening triage.

WhatsApp is the obvious channel for this in Indonesia. With approximately 112 million active users and a penetration rate of around 92% among Indonesian internet users as of Q3 2024, WhatsApp is the default messaging layer for virtually the entire addressable population of healthcare consumers.^[6] Unlike purpose-built health apps that require downloads, registrations, and behavioral adoption, a WhatsApp message to a hospital's verified number meets patients exactly where they already are — at 9 PM, on their phones, anxious about a symptom they cannot classify.

An AI pre-screening agent operating over WhatsApp can conduct a structured symptom assessment conversation in under two minutes: asking about the nature, duration, and severity of the presenting complaint; checking for red-flag symptoms that signal genuine emergency; assessing relevant history; and producing a routing recommendation with clear reasoning. Studies on AI-assisted emergency triage have shown agreement rates of 84–86% with physician assessment in structured triage environments,^[7] and the technology continues to improve as it is refined for specific clinical contexts.

The critical design principle: AI pre-screening does not diagnose. It routes. A pre-screening agent asks the right questions, applies a validated severity framework, and recommends the appropriate care pathway — with the explicit instruction to escalate any uncertainty to emergency care rather than under-triage. The safety architecture is conservative by design.

How Qlar's hc-triage Agent Works

Qlar's hc-triage agent is designed specifically for this pre-arrival screening use case. It is a conversational AI agent that operates over WhatsApp, conducting structured pre-screening for patients before they decide where to seek care. Here is how the interaction unfolds in practice:

• Symptom collection: The agent initiates a structured conversation to collect the patient's primary complaint, symptom duration, and severity. It asks targeted follow-up questions — not a generic checklist, but a branching conversation that adapts to what the patient reports. A patient reporting chest pain is asked different questions than one reporting lower back pain.
• Emergency detection and escalation: The agent continuously evaluates red-flag symptom combinations — chest pain with shortness of breath, stroke-like symptoms, signs of severe allergic reaction, altered consciousness, heavy bleeding — and immediately flags these for escalation to the triage nurse, with the patient's full symptom context transmitted before the call is even made. Emergency cases are never routed to self-care.
• Specialist routing for non-urgent cases: For patients whose symptom profile does not indicate emergency, the agent routes to the appropriate care level — recommending the right specialist type, identifying the earliest available appointment slot, and explaining why this pathway is appropriate for their presentation.
• Safe home monitoring guidance: For low-acuity presentations where monitoring at home is clinically safe, the agent provides clear, evidence-based instructions on what to watch for, when to seek care, and how to manage symptoms in the interim — turning a potential IGD visit into an educated self-care decision.
• Handoff with context: When a patient is referred to the IGD or a nurse, the handoff includes the full pre-screening summary — symptoms, duration, relevant history — so the clinical team receives context before the patient arrives, not after.

What hc-triage does not do is equally important: it does not prescribe medication, it does not diagnose conditions, and it does not override clinical judgment. It is a routing and information layer — designed to give patients the guidance they currently lack and give hospitals a filter that reduces non-urgent IGD volume without compromising the safety of patients who genuinely need emergency care.

Patient Categories and Routing Logic

Addressing the Safety Question Head-On

The most important objection to any AI triage tool is the safety question: what happens if the AI misroutes a serious case to outpatient care? This concern is legitimate, and the design response must be rigorous.

hc-triage is built with a conservative triage bias. When symptom data is ambiguous — when the combination of reported symptoms does not clearly exclude a serious etiology — the system defaults to escalation, not de-escalation. The protocol is: if in doubt, route to emergency. The cost of over-triage (directing a low-acuity patient to the IGD) is always lower than the cost of under-triage (directing a high-acuity patient to outpatient care). The agent is explicitly designed to err on the safe side.

Additionally, the agent does not operate in isolation. Every escalation and high-acuity routing generates a real-time notification to the hospital's clinical team. The system is a filter and router, not a gatekeeper — a patient who overrides the agent's recommendation and presents to the IGD anyway will be received and assessed by clinical staff. hc-triage reduces volume; it does not block access.

Research published in peer-reviewed journals confirms that AI-assisted triage systems can align closely with physician judgment while reducing assessment time. A 2025 systematic review and meta-analysis found that AI triage tools achieved an overall accuracy of 84.6% against clinical standard triage, with strong performance in identifying mid-range and high-acuity cases.^[7] The technology is not perfect — no triage system, human or AI, achieves 100% accuracy — but implemented with the right safety architecture, it meaningfully improves the match between patient acuity and care pathway.

Coming Soon: Qlar's Vision for Pre-Arrival Triage in Indonesia

hc-triage is currently in development as part of Qlar's healthcare AI platform. This article is intentionally forward-looking — we are describing where AI-powered pre-screening is going, not claiming it is fully deployed today. But the clinical case, the technical infrastructure, and the channel opportunity are all present. What is needed now is thoughtful implementation within the Indonesian healthcare context.

For hospital administrators and clinical directors reading this, the relevant question is not whether AI pre-screening will become part of healthcare delivery in Indonesia — it will — but whether your institution will be part of shaping how it is done responsibly. The hospitals that deploy this infrastructure early will accumulate clinical data, refine their routing protocols, and build patient trust in the channel. Those that wait will inherit solutions designed without their patient population in mind.

Qlar is inviting a limited number of hospitals and healthcare networks to participate in the early-access program for hc-triage. Early partners will work directly with our clinical and AI team to configure the agent for their specific patient profile, specialty mix, and IGD intake patterns — and will have input into how the routing logic is calibrated for the Indonesian triage context.

The Business Case: What 30% Fewer Non-Urgent IGD Visits Means in Practice

For a mid-sized Indonesian hospital receiving 200 IGD patients per day, if 30–40% of those visits are non-urgent — a conservative estimate based on the global and Indonesian data — that represents 60–80 patients per day who could be safely routed to alternative care pathways. Diverting even half of those patients through AI pre-screening has a cascade of operational effects:

• Reduced IGD wait times: Fewer patients in queue means faster initial assessment for genuinely urgent cases — a direct improvement in clinical outcomes.
• Staff workload reduction: Triage nurses and emergency physicians spend less time processing false emergency cases, allowing more focused attention on complex acute presentations.
• Outpatient conversion: Non-urgent patients routed to specialist outpatient clinics represent appointment revenue that is often more efficiently billed than emergency episodes under BPJS tariff structures.
• Patient satisfaction: Patients who are appropriately routed — avoiding a multi-hour IGD wait for a condition that could be managed at a clinic — consistently report higher satisfaction with the care experience.
• Data infrastructure: Every pre-screening interaction generates structured symptom data that, over time, allows hospitals to better understand their community's health patterns and adjust outpatient capacity accordingly.

Conclusion: The Fix Is Before the Door, Not Inside It

Indonesia's IGD overcrowding problem will not be solved by adding more beds, hiring more triage nurses, or expanding emergency capacity. It will be solved — or substantially mitigated — by changing when patients decide to come to the emergency department in the first place. That decision, made at home, at 9 PM, by someone who is anxious and underinformed, is exactly where AI can intervene.

A two-minute WhatsApp conversation with an AI pre-screening agent — one that asks the right questions, recognizes genuine emergencies, and routes intelligently across care pathways — is not a replacement for clinical care. It is the information infrastructure that Indonesia's healthcare system is currently missing. For the patient who would have spent four hours in an IGD waiting room for a mild fever, it is a faster path to appropriate care. For the patient in the middle of a cardiac event, it is a faster arrival with clinical context already in the hands of the team waiting for them.

Qlar's hc-triage agent is being built to close exactly this gap. For hospitals ready to be part of that future, early access is now open. The overcrowded IGD is a symptom of a solvable problem — and the solution starts before your next patient reaches the door.

Sources

• ^[1] Rosen P et al. “Overcrowding Indicators in Emergency Departments Across Countries: Scoping Review.” Interactive Journal of Medical Research, 2026. i-jmr.org/2026/1/e78073
• ^[2] Weinick RM et al. “National Study of Non-urgent Emergency Department Visits and Associated Resource Utilization.” Western Journal of Emergency Medicine, 2013. PMC3876304
• ^[3] Regional hospital study cited in: “GAMBARAN PENANGANAN PASIEN GAWAT DARURAT DI RUANG IGD RSUD.” Jurnal Kesehatan Perintis, 2023. South Kalimantan hospital data.
• ^[4] Shin TG et al. “Maximum emergency department overcrowding is correlated with occurrence of unexpected cardiac arrest.” Critical Care, BMC/Springer. PMC7276085
• ^[5] Assessing the Impact of External and Internal Factors on Emergency Department Overcrowding. International Journal of Emergency Medicine, Springer, 2026. doi.org/10.1186/s12245
• ^[6] Statista/Meltwater. “Indonesia social network penetration Q3 2024 — WhatsApp 92% penetration rate.” statista.com/statistics/284437/indonesia-social-network-penetration
• ^[7] Almutairi A et al. “Safety and accuracy of AI in triaging patients in the emergency department.” International Journal of Emergency Medicine, Springer, 2025. PMC12636208