• No se han encontrado resultados

Creación de librerías para secuenciación Ion Torrent ™

relacionados con los caracteres agronómicos de interés

CAPÍTULO 3: SECUENCIACIÓN DE AMPLICONES E IDENTIFICACIÓN DE PATRONES

3.2.3. Creación de librerías para secuenciación Ion Torrent ™

Aim 3 and Aim 4 focused on identifying inhalation anthrax outbreaks using a syndromic surveillance approach. Different scenarios were simulated and then outbreak detection methods were applied to understand how outbreak size, distribution, length, and start date affect the accuracy, timeliness, and rate of false alarms of each technique. Aim3 and Aim 4 are formally stated in Section 7.3 and Section 7.4.

These aims hold public health significance because of the syndromic surveillance has not focused on small, severe outbreaks. In particular no syndromic surveillance study has dealt with the detection of a small, deadly outbreak such as occurred in 2001, claiming the lives of 6 American citizens and hospitalizing 5 others. The results of the current study examine the 2001

attack in fine detail and answer the question “Would the 2001 attack be detected were it to occur at PUH?” The key findings of the research represent important first steps to reaching these goals.

Aim 3 identified three major disease clusters in the 2001 anthrax attack. In all 5 disease scenarios were identified to test the outbreak detection methods. For the two largest scenarios, the distribution of cases and start date was randomized. Although these simulations almost exhaustively present possible outbreak scenarios for testing, some limitations arise from certain design decisions. One limitation was due to the mean (0.37 cases/day) and standard deviation (0.61 cases) of the underlying time series. Because the mean was less than one case, it did not make sense to allow non-integer number of cases into the simulation. In addition, because the 2001 anthrax attack did not result in more than two patient admissions in one day, this limit was also used for the current study. Two excess cases in a single day represent an increase of 3.2 standard deviations of typical case volume.

Studies such as those conducted by Jackson et al. and Zhu et al. used up to five times the standard deviation to model case volume. (268,269) This was not justified in the current study where testing the actual cases counts was more important the hypothetical case counts. Although this simulation approach limits the generalizability of the findings for syndromic surveillance, it strengthens the findings for an outbreak specifically resembling the 2001 attack.

Aim 4 focused on identifying inhalation anthrax using 3 different outbreak detection methods in a variety of scenarios drawn from the 2001 anthrax attack. The first scenario superimposed the 2001 anthrax attack on to the 2001 PUH baseline identified in Aim 1. All three methods detected this outbreak on the first day. The success of each method may have been affected by an external factor. The baseline of syndromic cases was at an annual low from August 25th through the end of the year. At the time of the attack each method was forecasting

zero expected cases. This scarcity of syndromic activity does not appear to coincide with September 11th attack as it began two weeks prior. One way to better understand how the day of

the anthrax attack, October 16th, affected the accuracy of the methods was to vary the start date

of the attack. The “Cluster 1” simulation randomly chose start dates for the initial case cluster from the 2001 attack – 2 cases in 2 days. The outbreak detection performance declined once start date was varied from perfect detection across all three methods to 59% detection.

In terms of timeliness, the 2001 attack was detected on the first day by all three methods. When start date was randomized in “Cluster 1” simulations, only 20-30% of outbreaks were detected on the first day.

During this historically low period of syndromic activity, false alarms also increased. From September 13th to November 30th, every day with a syndromic case caused a false alarm –

meaning the outbreak detection method registered a significant increase greater than 0.01 probability. This is not unexpected with syndromic surveillance. Timely systems tradeoff a certain number of false alarms for the assurance outbreaks are detected as quickly as possible.

Aim 4 uncovered trends in the minimum size of an outbreak able to be detected by the three methods. A clear trend is shown by looking at the accuracy of Cluster 3 (43% outbreaks detected), Cluster 1 (59% of outbreaks detected) and Cluster 2 (97% of outbreaks detected). This may also be affected by the length of the outbreak. The Cluster 3 had only one day to be detected within the outbreak interval. In many cases the outbreak occurred before the increase had a chance to be registered. Cluster 2, which is up to 11 days long, allows greater time for the increase to develop. For instance, the Z method detected 97% of Cluster 2 outbreaks. Approximately, 60% of those were detected after the second day.

These results are consistent with the findings of Murphy who found 60% detection of “Slow-building” outbreaks using the same z-score methodology. However, the present study greatly improves on the results in the Jackson study looking at C1 and C3 performance for small outbreaks. (See Section 5.5) In the study only 10% of outbreaks under 5 cases were detected by the two methods. This can be explained by the mean and standard deviation of the underlying baselines chosen for the study. The lowest volume was for pneumonia hospitalizations with a mean of 2 cases per day with a SD of 1.6 cases per day with the highest volume represented by a general respiratory syndrome with a mean of 60 cases per day and a SD of 16. Because Murphy presents his findings in aggregate, it is impossible to tell the C1 and C3 performance on just the pneumonia time series. This study affirms C1 and C3 would far exceed only 10% performance given a lower volume baseline.