To use Internet search data to compare duration of compliance for various diets.

Using a passive surveillance digital epidemiological approach, we estimated the average duration of diet compliance by examining monthly Internet searches for recipes related to popular diets. We fit a mathematical model to these data to estimate the time spent on a diet by new January dieters (NJD) and to estimate the percentage of dieters dropping out during the American winter holiday season between Thanksgiving and the end of December.

Internet searches in the USA for recipes related to popular diets over a 15-year period from 2004 to 2019.

Individuals in the USA performing Internet searches for recipes related to popular diets.

All diets exhibited significant seasonality in recipe-related Internet searches, with sharp spikes every January followed by a decline in the number of searches and a further decline in the winter holiday season. The Paleo diet had the longest average compliance times among NJD (5.32 ± 0.68 weeks) and the lowest dropout during the winter holiday season (only 14 ± 3 % dropping out in December). The South Beach diet had the shortest compliance time among NJD (3.12 ± 0.64 weeks) and the highest dropout during the holiday season (33 ± 7 % dropping out in December).

The current study is the first of its kind to use passive surveillance data to compare the duration of adherence with different diets and underscores the potential usefulness of digital epidemiological approaches to understanding health behaviours.

Hurricane Sandy's October 29, 2012 arrival in New York City caused flooding, power disruption, and population displacement. Infectious disease risk may have been affected by floodwater exposure, residence in emergency shelters, overcrowding, and lack of refrigeration or heating. For 42 reportable diseases that could have been affected by hurricane-related exposures, we developed methods to assess whether hurricane-affected areas had higher disease incidence than other areas of NYC.

We identified post-hurricane cases as confirmed, probable, or suspected cases with onset or diagnosis between October 30 and November 26 that were reported via routine passive surveillance. Pre-hurricane cases for the same 4-week period were identified in 5 prior years, 2007–2011. Cases were geocoded to the census tract of residence. Using data compiled by the NYC Office of Emergency Management, we determined (1) the proportion of the population in each census tract living in a flooded block and (2) the subset of flooded tracts severely “impacted”, e.g., by prolonged service outages or physical damage. A separate multivariable regression model was constructed for each disease, modeling the outcome of case counts using a negative binomial distribution. Independent variables were: neighborhood poverty; whether cases were pre- or post-hurricane (time); the proportion of the population flooded in impacted and not impacted tracts; and interaction terms between the flood/impact variables and time. Models used repeated measures to adjust for correlated observations from the same tract and an offset term of the log of the population size. Sensitivity analyses assessed the effects of case count fluctuations and accounted for variations in reporting volume by using an offset term of the log of total cases.

Only legionellosis was statistically significantly associated with increased occurrence in flooded/impacted areas post-hurricane, adjusting for baseline differences (P = .04). However, there was only 1 legionellosis case post-hurricane in a flooded/impacted area.

Hurricane Sandy did not appear to elevate reportable disease incidence in NYC. Defining and acquiring reliable data and meta-data regarding hurricane-affected areas was a challenge in the weeks post-storm. Relevant metrics could be developed during disaster preparedness planning. These methods to detect excess disease can be adapted for future emergencies. (Disaster Med Public Health Preparedness. 2013;7:513-521)

The number of civilians killed in Iraq following the 2003 invasion has proven difficult to measure and contentious in recent years. The release of the Wikileaks War Logs (WL) has created the potential to conduct a sensitivity analysis of the commonly-cited Iraq Body Count's (IBC's) tally, which is based on press, government, and other public sources.

The 66,000 deaths reported in the Wikileaks War Logs are mostly the same events as those previously reported in the press and elsewhere as tallied by iraqbodycount.org.

A systematic random sample of 2500 violent fatal War Log incidents was selected and evaluated to determine whether these incidents were also found in IBC's press-based listing. Each selected event was ranked on a scale of 0 (no match present) to 3 (almost certainly matched) with regard to the likelihood it was listed in the IBC database.

Of the two thousand four hundred and nine War Log records, 488 (23.8%) were found to have likely matches in IBC records. Events that killed more people were far more likely to appear in both datasets, with 94.1% of events in which ≥20 people were killed being likely matches, as compared with 17.4% of singleton killings. Because of this skew towards the recording of large events in both datasets, it is estimated that 2035 (46.3%) of the 4394 deaths reported in the Wikileaks War Logs had been previously reported in IBC.

Passive surveillance systems, widely seen as incomplete, may also be selective in the types of events detected in times of armed conflict. Bombings and other events during which many people are killed, and events in less violent areas, appear to be detected far more often, creating a skewed image of the mortality profile in Iraq. Members of the press and researchers should be hesitant to draw conclusions about the nature or extent of violence from passive surveillance systems of low or unknown sensitivity.

CarpenterD, FullerT, RobertsL. WikiLeaks and Iraq Body Count: the Sum of Parts May Not Add Up to the Whole—A Comparison of Two Tallies of Iraqi Civilian Deaths. Prehosp Disaster Med. 2013;28(3):1-7.