Up-to-date information is essential to public health decision-making. However, reporting delays can pose challenges to determining recent trends. Nowcasting methods can help address this challenge.
Different types of qualitative assessments and modeling outputs are useful at different horizons.
Nowcasts estimate real-time disease burden from partially reported data by adjusting for historical reporting patterns -improving situational awareness and aiding decision-making.
Nowcasts estimate current disease burden metrics, such as numbers of hospitalizations or emergency department visits, by adjusting for delays in reporting. All disease tracking systems are subject to some lag in data reporting. This is accounted for by taking snapshots from surveillance data to measure the delay between when an event occurs and when it is reported. Nowcasts are most accurate for signals with stable, predictable reporting delays, like seasonal pathogens, and may be more challenging in early outbreak situations.
Nowcasting is especially helpful when there is a reporting lag that might otherwise make real-time estimates look misleading. Here is an example of how nowcasting can be used: with incoming data, it can be challenging to know whether declines in recently reported data are due to true declines or to delays in reporting. You might see this first batch of data and assume the number of emergency department visits are decreasing, but in reality - with time - you will see they are not.
The next day, more reports come in as a second data batch and the data from the past few days are updated.
The following day, even more reports come in, and the trend now appears to be increasing.
In this instance, the nowcast can be made even without the additional data batches, and it can correctly adjust for this partial reporting by "filling in" the incomplete data based on historic reporting patterns, to provide better situational awareness in real-time. For example, if we know that only 30% of emergency department visits are reported on the same day as the actual visit, we can estimate the final number of emergency department visits for that day by adding the expected 70% that have not yet been reported. Nowcasting models also provide an estimate of uncertainty in the final reported counts through prediction intervals.
Nowcasts typically rely on stable surveillance data and well-understood reporting delays. Trends may not be straightforward and delays can vary over time. This is especially true during outbreaks, when the pattern of reporting changes often.
Reporting from different locations or health facilities can come in at different times and with varied completeness. Adjustments can be made to try and account for these challenges. Weekend and holiday reporting delays are some of the most commonly incorporated adjustments. When there is uncertainty, this is reflected in wider prediction intervals.
Data collectors and managers should remain in close contact with modelers, especially in an outbreak context, to ensure that the nowcasts are adjusted to the latest data environment.
To summarize, nowcasting produces real-time estimates based on currently available but incomplete reported data and historical reporting patterns. It can potentially reveal changes in disease transmission dynamics before they would otherwise be detected. You can use many different metrics for nowcasting, such as case counts, emergency department visits, hospitalizations, and deaths.
Nowcasts can help provide a picture of current disease trends while there are data reporting lags or gaps.