Congo Is Counting Fewer Than Half Its Ebola Cases

Estimating The Invisible

In mid-July the head of WHO’s emergencies program, Chikwe Ihekweazu, told reporters that “80% of new cases are outside our contact lists and so are coming to us from unknown chains of transmission.” The figure is a clue to the outbreak’s true size. It is also a slippery one.

The eighty percent confounds two quantities: the cases absent from a contact list when symptoms begin, and the cases with no identifiable source after investigators have finished with them. Only the second says anything about how many infections ultimately go uncounted. A patient can be a complete surprise to the surveillance system on Monday and have a fully reconstructed chain of transmission by Friday.

An email to me from WHO provides the information needed to resolve the ambiguity. Currently, between 40 and 45 percent of cases have a known epidemiological link once investigators are done, and the 20 to 25 percent who come directly from contact lists are a subset of those. The gap, fifteen to twenty-five percentage points, is patients nobody was watching whose source could be worked out afterwards. The remaining 55 to 60 percent has no identifiable source at all.

The agency also gave me the tally underneath. In July, 309 of 671 confirmed cases documented well enough to judge had a known link, or 46 percent, a proportion WHO says has held steady since early June. That implies a little more than two infections for every one counted.

Counting What Was Never Counted

Case investigators can only trace a patient back to whoever infected them if that person was caught by surveillance too. So the fraction of cases with a known infector doubles as a rough measure of how many infectors the system was finding in the first place. That is the detection rate, and dividing the confirmed count by it gives an estimate of the true number of cases.

At 46 percent, the 3,822 confirmed cases imply about 8,300 infections. I would consider that a lower bound. It is a very soft one, because the 46 percent was measured only on cases documented well enough to judge, and the sparse ones left out are precisely those least likely to have a traceable source.

The 20 percent already on a contact list sets the ceiling. Anyone on a list before they fall ill is nearly certain to be counted, so that figure understates detection. The fivefold correction it implies, around 19,000 infections, is about as far above the count as the truth can plausibly sit.

The two bounds are not symmetrical. The ceiling is firm. The floor is soft, and the way it is calculated all but guarantees the truth lies above it. WHO’s three-to-four-fold estimate sits inside the bracket, above the middle.

Publicly the agency has been more cautious. Tedros Adhanom Ghebreyesus, the director-general, has warned that the true number of cases could be “more than double” the official count. The figure the agency sent me is roughly twice that again. A European Centre for Disease Prevention and Control review from June went wider still, citing a model that put infections at 3.0 to 10.2 times reported. That interval is too wide to plan around. It also does not contain the possibility that the official count is roughly right.

The estimate running lower is a nowcast maintained at the London School of Hygiene and Tropical Medicine, which puts total infections between 5,700 and 11,200 with 90 percent probability. WHO cited that group as consistent with its own number, and at the end of July it was. The two have drifted apart since, for reasons that have less to do with the July data than with what the model assumes about January. That is a separate story.

How Many People Have Died?

The correction that scales confirmed cases up to true infections applies to infections. It does not transfer cleanly to estimate the death toll. A body is harder to overlook than a mild illness, and the infections surveillance misses are weighted toward the milder ones, in people who never reached a treatment center. The real toll is above the official 1,707, but probably by less than three to four fold. Deaths also lag infections by weeks: the LSHTM model estimates between 246 and 608 further deaths among people infected before August 1, even if transmission stopped tomorrow. But, of course, it won’t.

The share of confirmed patients who die makes the same point from the other end. That figure stood at 28.8 percent at the end of June. A month later it had reached 44.1 percent, a rise WHO attributes to “persistent delays in case detection, referral, and access to treatment, together with the continued predominance of community deaths.” A surveillance system that finds people late, or only once they have died at home, will report a death rate higher than the disease itself carries, because the patients it never finds are disproportionately the ones who lived.

The Number Nobody Will Ever Measure

Every number here is an inference drawn from the behavior of the surveillance system rather than a direct measurement of the epidemic. A serological survey, or the systematic testing of community deaths that followed the West African epidemic, would replace the inference with a count.

Neither is going to happen. Not across 49 health zones in five provinces, not in the middle of an active emergency, and not where WHO reports that “insecurity and attacks affecting health facilities have hampered response operations.” Every decision must be made against a number nobody knows for sure.

That puts the weight of decision-making on the trajectory of the epidemic. The nowcast currently puts the reproduction number, the average number of people each patient goes on to infect, at about 1.1, with uncertainty running to either side of one. One might be encouraged that it is down from its original value between 2 and 3. It is not a reason for complacence. An epidemic at 1.1 is still growing, and one at 0.95 shrinks so slowly it could run for another year. Ending this in months rather than years will require something nearer 0.5, which means cutting transmission by more than half from the current rate, at the same time, across the 33 health zones where it is still active.

A much larger outbreak is still possible. WHO rates the risk inside Congo as “very high.” The epidemic is already urban: the largest single cluster is Bunia, the provincial capital of Ituri, with 880 confirmed cases. It is moving through a region of armed groups and displaced populations, and 151 health workers have been infected. West Africa in 2014 was a rural outbreak until it reached the capitals, and it ended with more than 28,600 cases and 11,300 deaths. Nothing in the present trajectory rules that out.

Nobody is going to go and count. The outbreak on paper stays what it has been since January, a shadow of the real one, and the response has to be built for the epidemic casting that shadow.