Recent seroprevalence survey in New York City points to 35.6% infected and a 0.9% IFR


News came a few hours ago of a serology survey in New York City this week. We matched our model on reported data and found an ultimate Infection Fatality Rate of 0.9% for New York City, significantly higher than the 0.21% assumed in our previous model on 15 April – which was derived from the Denmark model. This preliminary IFR figure will need to be confirmed with more advanced epidemic models. The aim of this quick article is mainly to illustrate how important it is to consider epidemics dynamics when interpreting seroprevalence surveys. If confirmed by the scientific community, the high IFR, consecutive to a rather low observed seroprevalence percentage, will indicate a disappointingly low level of acquired herd immunity. We confirm that the value of information of well planned serology surveys is enormous, potentially saving thousands of lives if they contribute defining and implementing the right policies. This study, again, points to contrasting results from country to country, which draws two comments: first, it is one more reason to continue accumulating these results impartially; second, one should not necessary be in search for one single universal IFR number.


We assume: New York City, seroprevalence 21.2% on median date 21 April 2020 (“this week”).


We use CovModel 1.5, a simple SIR model. The total number of infected and removed is linked to the death curve by a proportion factor (the IFR) and a time lag (17 days). the IFR itslef is the product of a fixed paramter, the ratio of death to severe cases (0.14) and the ratio of severe to total (rho). Hence, assuming rho or the IFR is strictly equivalent. IFR = 0.14 x rho. We simulated seroprevalence from offsetting infection curve by 15.4 days, as explained in a recent article.

We simultaneously matched death rate curve and the observed seroprevalence point on 21 April, by varying, by trial and error:

  • rho (or IFR) to provide the correct seroprevalence
  • Ro to provide the correct slope of death curve
  • Time of first infection to provide the right date for first death
  • R schedule with time to model social distancing and get the shape of death rate


Figure 1 shows the matched death curve and Figure 2 the matched seroprevalence curve.

Figure 2 illustrates seroprevalence time lag compared to infection curve, illustrating how important it is to take dynamics into account when interpreting serology surveys. Epidemics models such as CovModel, and if possible more advanced SEIR multi-age and multi-symptoms models, are a necessity to interpret seroprevalence.

Figure 3 shows that the inferred number of infected from simulation is at least an order of magnitude higher than the official confirmed incidence, even though the IFR is particularly high.

Total Confirmed Cases in New York City was 147,297 on 22 April. Simulated total infected or removed is almost exactly 3 millions (ratio 1 in 20). On 21 April, a 21.2% seroprevalence rate translates into a 35.6% infected rate according to our model.

Figure 1: Death curve match. Simulated susceptible population suggest herd immunity not achieved, since about 50% of population could remain not infected
Figure 2: simulated seroprevalence curve matched to NYC observation on 21 April. In red: simulated total infectious + removed (recovered or dead). This plot demonstrates the epidemic dynamics and time lag between seroprevalence and infection.
Figure 3: death rate match (black). Death records changed accounting standard on 16 April. Our model tries to incorporate the additonal death count. Red curves are simulated cases incidence and recorded Confirmed Cases incidence.