When Labs Lie: The Art of Spotting Fake Disease Spikes
I spent three weeks chasing what I thought was a novel respiratory virus moving through a nursing home. Twenty-seven positive PCR results in ten days. The county health department was already drafting press releases when I noticed something odd. The cycle threshold values were climbing. Day one samples averaged 28 cycles. By day eight, they were sitting at 33. Real infections don't gradually weaken like that. We pulled the lab records, checked the equipment logs, and found a reagent lot issue that had been contaminating everything since Tuesday. That was a pseudo-outbreak, and it cost us forty hours of investigative work before we could prove nothing was actually spreading. A pseudo-outbreak happens when surveillance data suggests an increase in disease cases, but the signal comes from artifact rather than actual transmission. The numbers look real. The line graphs spike the way they should. Something feels wrong underneath. These false alarms show up everywhere now because we're testing more aggressively and reporting faster. The problem isn't the data quality itself. It's that our interpretation hasn't caught up with the speed of collection. The most common causes fall into three buckets. Laboratory contamination dominates, usually from cross-sample mixing or reagent degradation. Changes in testing protocol create the second category. Someone decides to switch from culture to PCR, or starts testing asymptomatic patients, and suddenly the case count doubles overnight. Reporting behavior changes round out the list. A new electronic health record system auto-populates fields that previously went blank, or a clinic starts faxing results when they used to call them in. All three produce the same visual pattern: a sudden vertical move on the epidemic curve that looks terrifying until you dig into the metadata.
I learned to approach every spike the same way. First, I check the test methodology. What assay is being used, what specimen type, and has anything changed in the past thirty days? Second, I pull the cycle threshold values or colony counts depending on the organism. Real outbreaks cluster in a narrow range. Artifacts spread widely or trend downward over time. Third, I examine the patient population. Are all the cases coming from one unit, one shift, one lab technician? If yes, the signal is probably environmental rather than biological. Here's the counter-intuitive part that trips people up. Sometimes pseudo-outbreaks reveal real problems masked as false alarms. The contaminated reagent might indicate poor storage practices. The reporting change might expose systematic undercounting that was hiding a genuine outbreak. I once saw a hospital that switched to a newer rapid influenza test and immediately reported a thirty percent increase in cases. The test was less sensitive, not more. What looked like an outbreak was actually worse understaffing and delayed testing. The pseudo-outbreak uncovered the real issue. Advanced investigators look at the denominator, not just the numerator. Case counts mean nothing without knowing how many tests were run, how many patients were screened, and what the baseline rate was. I calculate attack rates per hundred tests, not per hundred patients. This catches spikes that look dramatic but are actually proportional to increased testing volume. A fifty percent increase in positive results sounds alarming until you realize testing volume went up sixty percent. The rate actually dropped.
Time series decomposition helps separate signal from noise. You can decompose a surveillance curve into trend, seasonal, and irregular components. Pseudo-outbreaks usually show up in the irregular component as isolated spikes without the sustained elevation you'd expect from true transmission. I use the STL method, seasonal-trend decomposition using LOESS, because it handles missing data better than older approaches. The irregular component tells you what's left after removing the predictable patterns. If that residue spikes sharply and returns quickly, you're probably looking at artifact. Geographic clustering analysis catches another common failure mode. Real outbreaks spread through contact networks. Pseudo-outbreaks concentrate around testing sites or reporting pathways. I map cases by specimen collection location, not just patient residence. When five positive MRSA cultures all came from the same phlebotomy cart, I knew the bacteria wasn't spreading through the ward. The cart was contaminated. We replaced it, ran environmental cultures, and the "outbreak" stopped after forty-eight hours without a single case of actual transmission. The workaround I use for persistent lab contamination issues involves batch analysis. Most labs run specimens in groups of twenty to fifty. If you see positive results trending from batch to batch, with the first positive appearing around the middle of one batch and the last at the beginning of the next, the contamination is happening during processing. I flag these patterns immediately and request independent confirmatory testing before reporting anything to public health. This usually saves twelve to eighteen hours of investigation and prevents unnecessary isolation measures.
Get the Full Details
Here's what nobody tells you about pseudo-outbreaks. They're harder to dismiss than real outbreaks. With a real outbreak, the evidence accumulates. More cases, longer duration, clear transmission chains. With a pseudo-outbreak, you have to prove a negative. You have to show why the signal isn't real, which means documenting every alternative explanation until the simplest one remains. I spend more time investigating false alarms than actual epidemics, and I'm still not good at it. The skill comes from pattern recognition built over years of wrong calls. Sometimes the best investigation is no investigation. When a single positive result appears in an asymptomatic patient with no exposure risk, I don't launch an outbreak response. I verify the lab result, check the specimen collection method, and document the finding. If the repeat test is negative, I close the chart and move on. This happens maybe twice a week in my practice. The temptation to overreact is strong, especially when leadership is watching the surveillance dashboard. Resist it. False positives generate false urgency, and false urgency wastes resources that real outbreaks need. Documentation matters more than you'd think. When you conclude a pseudo-outbreak, write it down. Not just the conclusion, but the reasoning. What pattern did you observe, what alternative explanations did you consider, why did you reject them? Future investigators will thank you. I've reviewed pseudo-outbreak investigations from three years ago and found my own notes still useful because I'd written the decision tree, not just the outcome. This practice also protects you when someone questions why you didn't escalate. Documentation shows due diligence.
Automated detection systems are getting better but still miss the subtle patterns humans catch. I've tested commercial algorithms against my own investigations, and they flagged seventy percent of true outbreaks correctly but also generated forty percent false positives. The tradeoff is acceptable for population-level surveillance but dangerous for individual facility responses. I use automated systems as triage tools, not decision tools. When the algorithm says "possible outbreak," I investigate. When it says "normal," I still check the raw data because algorithms trained on historical patterns sometimes miss novel contamination routes or unusual reporting artifacts. The most frustrating pseudo-outbreaks involve diagnostic drift. A lab upgrades its equipment, changes its reference ranges, or modifies its interpretation criteria without telling the clinicians. Case counts change overnight because the definition of positive changed, not because disease patterns changed. I learned this the hard way when our hospital switched to a new streptococcus testing platform and immediately reported a twenty-two percent increase in pharyngitis cases. The new test detected carriers that the old test missed. Were those real cases or just colonization? The answer depends on your clinical definition, which varies by specialty and season. Regulatory reporting requirements sometimes incentivize pseudo-outbreak investigations. Joint Commission surveys, CDC NHSN reporting, state notifiable disease programs. When compliance depends on documenting outbreak responses, there's pressure to find something even when the evidence is thin. I've seen nurses extend observation periods and physicians order additional cultures because the dashboard showed a borderline elevation. This creates self-fulfilling prophecies. The investigation generates more data, the data looks suspicious, the suspicion justifies continued investigation. Breaking this cycle requires explicit criteria for when to stop looking, not just when to start.
Communication strategy differs for pseudo-outbreaks versus real outbreaks. With real outbreaks, you share information broadly and frequently. With pseudo-outbreaks, you share minimally and selectively. Announcing a false alarm to staff creates anxiety and undermines trust in the surveillance system. I notify the infection prevention team, the lab director, and the epidemiologist. Then I document the conclusion and archive the investigation. If leadership asks why I'm not sending hospital-wide alerts, I explain that the data shows artifact, not transmission, and that unnecessary alarm disrupts care more than silence. This conversation happens regularly, and I still find it uncomfortable. Statistical process control charts catch pseudo-outbreaks that case counts miss. I plot daily or weekly test positivity rates with control limits set at three standard deviations from the mean. Points outside the limits trigger investigation, but so do runs of seven points trending in one direction, even if they stay within the limits. This catches gradual contamination or reporting drift that single-point spikes don't. The method comes from industrial quality control, adapted for epidemiology in the nineties. It works because the underlying math doesn't care whether you're monitoring bolt thickness or bacterial counts. Both are just measurements subject to variation. Seasonal adjustments complicate everything. Respiratory viruses peak in winter. Gastrointestinal outbreaks surge in summer. Allergy testing spikes in spring. If you compare June case counts to May without accounting for seasonality, you'll misinterpret normal variation as abnormal signal. I calculate seasonally adjusted rates using moving averages or regression residuals, depending on the pathogen and data availability. This adds twenty minutes to my weekly review but prevents maybe two false alarms per year. The time investment pays for itself.
Here's a specific edge case I encountered last winter. A pediatric clinic reported five positive RSV results in seven days, up from their usual two per week. The spike looked suspicious because it coincided with a staff change. The new lab technician started reporting results the same day, suggesting a possible contamination event. But the specimens came from different patients, different days, different collection methods. I requested independent testing, and all five results confirmed. We investigated further and found that the clinic had switched to a newer rapid antigen test two weeks earlier. The new test was more sensitive, detecting milder cases that the old test missed. The "outbreak" was real, but the increase was due to improved detection, not increased transmission. This distinction matters for public health response. More sensitive testing doesn't require more isolation. It requires better communication about what the numbers actually mean. The workaround for this scenario involved baseline recalibration. Instead of comparing current counts to historical averages calculated with the old test, I recalculated baseline using the first thirty days of new-test data. The adjusted rate showed no significant increase. This approach takes extra time upfront but prevents unnecessary interventions downstream. I apply it whenever testing methodology changes mid-surveillance, regardless of whether the change appears to cause spikes. The principle is simple: you can't compare apples to oranges, even if both are fruit. False alarms teach you more than true outbreaks. When you investigate a pseudo-outbreak, you learn the weaknesses in your surveillance system, the blind spots in your data collection, the assumptions you've never questioned. I keep a running log of every false alarm I've encountered, categorized by cause and resolution method. After five years, the log has forty-three entries. About forty percent involved laboratory issues, twenty-five percent reporting changes, fifteen percent statistical artifacts, and the remaining twenty percent fell into miscellaneous categories I hadn't anticipated. Having this history helps me recognize patterns faster and dismiss unlikely scenarios sooner.
Cost analysis shows pseudo-outbreak investigations consume disproportionate resources. The average investigation takes four to six hours of staff time, plus any additional testing, isolation, or communication required. Multiply that by the number of false alarms per year, and you're looking at significant operational burden. I estimate my facility spends roughly one hundred twenty hours annually investigating pseudo-outbreaks. That's three work weeks diverted from other priorities. The question isn't whether we can eliminate false alarms. The question is whether we can investigate them more efficiently while maintaining sensitivity to true outbreaks. Training reduces both false positives and false negatives. I run quarterly simulation exercises where staff practice responding to synthetic outbreak scenarios, some real, some pseudo. After each exercise, we review the decisions made and the evidence considered. This builds pattern recognition faster than reading manuals. Staff who complete six or more simulations per year make fewer classification errors than those who complete fewer than three. The training takes ninety minutes per session, but the return on investment shows up in reduced investigation time and improved response accuracy. Technology helps but doesn't solve the fundamental problem of interpretation. Automated surveillance systems generate alerts based on thresholds and trends. Humans decide what the alerts mean. The gap between alert and interpretation is where pseudo-outbreaks get misclassified. I've seen systems flag normal seasonal variation as outbreak because the threshold was set too low. I've also seen systems miss real outbreaks because the signal was subtle and buried in noise. Neither failure mode is acceptable, but both are inevitable when algorithms replace judgment. The solution isn't better algorithms. It's better-trained humans working alongside the algorithms.
Mortality and morbidity implications depend on the pathogen and population. A pseudo-outbreak of MRSA in a surgical ICU might trigger unnecessary contact precautions and bed closures. A pseudo-outbreak of influenza in a nursing home might generate panic and premature discharge decisions. I assess the potential harm of both overreaction and underreaction before launching any investigation. When the harm of missing a real outbreak outweighs the harm of investigating a false one, I err on the side of caution. When the resources required for investigation would compromise care for other patients, I limit the scope. This balancing act is the hardest part of the job, and there's no algorithm for it. Legal and liability considerations sometimes drive pseudo-outbreak responses. When a facility faces potential litigation, there's pressure to investigate everything thoroughly, even when the evidence suggests artifact. I've been asked to "cover all bases" by administrators worried about lawsuits. My response is consistent. I investigate based on evidence, not fear. If the data supports a pseudo-outbreak conclusion, I document it clearly and stand by it. If someone wants to investigate further despite the evidence, that's their decision, but I won't endorse it professionally. This stance creates friction occasionally, but it preserves integrity, which matters more than short-term convenience. The future of pseudo-outbreak detection involves multi-source data integration. Instead of relying on single surveillance streams, I'm combining laboratory data, pharmacy records, staffing patterns, environmental monitoring, and patient flow metrics. When all sources align, the signal is stronger. When they diverge, the divergence itself is informative. This approach takes more time and coordination but reduces both false positives and false negatives. I expect it to become standard practice within five years as data infrastructure improves and analytical methods mature. Until then, the fundamentals remain the same: question the data, understand the context, and let the evidence guide the conclusion.
