











Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Community
Ask the community for help and clear up your study doubts
Discover the best universities in your country according to Docsity users
Free resources
Download our free guides on studying techniques, anxiety management strategies, and thesis advice from Docsity tutors
Table 1 shows the stark rise in deaths due to Clostridium difficile en- terocolitis (ICD-10 A04.7) over a 14-year period. The United States began.
Typology: Exercises
1 / 19
This page cannot be seen from the preview
Don't miss anything!
Abstract Enterocolitis due to Clostridium difficile is major emerging cause of death in the US. Between 1999 and 2012, C. diff. deaths rose by a staggering almost-10-fold increase, to 7,739 from 793. This paper has three goals. First, we present a demographic description of C. diff. mortality in the US since 1999. Second, we test a hypothesis that the increase in C. diff. deaths is due to population aging. We find that the emergence of this cause of death follows a proportional hazard pattern, above age
Clostridium difficile is a gram-positive spore-forming bacterium, and a clinically- significant enteric pathogen (Sunenshine and McDonald 2006). Clinical C. diff. infection is almost always associated with antibiotic use (Bartlett 2008 a , Leffler and Lamont 2015). The Centers for Disease Control and Prevention (CDC) has identified C. diff. as one of three microoganisms at the highest ∗Final author version. Published as Biodemography and Social Biology 62(2):198–207 (2016). http://www.tandfonline.com/doi/full/10.1080/19485565.2016.1172957 † To whom correspondence should be addressed: noymer@uci.edu
Table 1: Number of deaths: C. diff. infection (ICD10 A04.7), US, 1999–
Number of deaths year underlying contributory total 1999 793 752 1, 2000 1,101 919 2, 2001 1,332 988 2, 2002 2,195 1,331 3, 2003 2,776 3,601 6, 2004 4,062 3,992 8, 2005 5,332 3,204 8, 2006 6,225 3,582 9, 2007 6,372 3,507 9, 2008 7,476 4,103 11, 2009 7,251 4,089 11, 2010 7,298 4,198 11, 2011 8,085 4,616 12, 2012 7,739 4,850 12,
threat level, “urgent”, along with carbapenem-resistant Enterobacteriaceae (CRE) and drug-resistant Neisseria gonorrhoeae , noting: “Although C. difficile is not currently significantly resistant to antibiotics used to treat it, it was included in the threat assessment because of its unique relationship with resistance issues, antibiotic use, and its high morbidity and mortality.” (Na- tional Center for Emerging Zoonotic and Infectious Diseases 2013).
Table 1 shows the stark rise in deaths due to Clostridium difficile en- terocolitis (ICD-10 A04.7) over a 14-year period. The United States began using ICD-10 for death codes in 1999, before which it is difficult to track C. diff. mortality. The reason is that in the ICD-9 coding scheme, C. diff. deaths were coded as 008.45 (5-digit code), whereas the mortality detail files report only 4-digit ICD-9 codes. For example, in 1998 there were 600 deaths in the US with cause of death ICD-9 008.4, “intestinal infections due to other spec-
rates do not change. We know, a priori , that a nearly 10-fold increase in deaths in 14 years (cf. table 1) must have a causal component beyond pop- ulation aging, but nonetheless it is desirable to describe the age-mortality profile for C. diff. , and whether it has changed over time.
We present underlying-cause and any-mention mortality rates for C. diff. infection, by age (5-year groups) and sex, calculated from data on every death in the US. We use cause-specific Gompertz models to test a propor- tional hazard hypothesis, and we introduce therein a weighted least squares technique for estimating Gompertz survival curves.
Clostridium difficile mortality
We used data from the multiple cause of death files of the National Center for Health Statistics (National Center for Health Statistics 2014), a database of all mortality in the United States. We extracted data on all deaths contain- ing C. diff. infection (ICD10 A04.7) as an underlying or contributory cause, during the period 1999–2012. The start of the time span corresponds to the beginning of ICD-10 death coding in the United States, and the end of the span reflects the most recent available data. Using population data from the Human Mortality Database (2014), we calculated age-specific and age- adjusted death rates for C. diff. infection, separately by sex. The year- United States standard population was used for the calculation of the age- adjusted rates. We also calculated all-mention death rates, i.e., where the numerators were all deaths involving C. diff. infection as an underlying or contributory cause (Wing and Manton 1981).
any occurrenceunderlying cause
males females
Figure 1: Age-adjusted death rate, C. diff. infection, 1999–2012. Figure 1 shows the age-adjusted death rate for C. diff. infection, 1999–2012. This graph shows three important features of C. diff. mortal- ity in the US. First, C. diff. age-adjusted death rates have risen about 7-fold since 1999, from less than 0.5 per 100,000 to just under 2.5 per 100,000. As table 1 shows, this corresponds to a 10-fold increase in the absolute number of deaths in which C. diff. is the underlying cause, and an 8-fold increase in all deaths involving C. diff. , in just a fourteen-year time span. The in- crease over time has been uneven, with a plateau from 2008–12. Second, there is no meaningful sex difference in the age-adjusted death rates for the underlying cause of death. Third, the all-mention age-adjusted death rates follow the underlying age-adjusted death rates over time in a roughly par- allel fashion, but since 2003, a sex difference has emerged. Specifically, in
males females
Figure 2: Age-mortality profile, C. diff. infection (underlying cause), 1999–2012.
male and female hazards often behave differently (Kohler and Kohler 2000), although in this case there is no qualitative distinction between the shapes or levels of the graph by sex, above age 40 (cf. figure 2). Five-year age groups, 40 ≤ x ≤ 99 (i.e., 40–44, 45–49,... ), were used to smooth age heaping. In each age group, the mean age of C. diff. mortality in the interval was used as the x value in the regressions. For example, for females age 65–69 in the more recent time period, the mean age of death due to C. diff. was 67.153, so this value was used as x instead of the midpoint of the interval, 67.5. This “centering” of the x values is used to polish the estimates, although it has small effects. Such centering, combined with parametric estimation of mortality, can help avoid some of the methodological pitfalls outlined by Gelman and Auerbach (2016).
Table 2: Proportional hazard analysis of C. diff. mortality, US, 1999–2005 vs. 2006–
WLS FEMALES MALES (1) (2) (3) (1) (2) (3) Early Late Early Late (1999–2005) (2006–12) ∆ (1999–2005) (2006–12) ∆ age term ( β ) 0.124*** 0.126*** 0.129*** 0.132*** (30.7) (26.8) (38.2) (43.2) intercept ( α ) −19.85*** −18.61*** −20.16*** −19.08*** (−59.8) (−48.0) (−75.6) (−78.0) period term ( γ ) 1.240 1. (1.80) (2.34) age×period ( δ ) 0.00157 0. (0.19) (0.56) N 12 12 24 12 12 24 R^2 0.9895 0.9863 0.9890 0.9932 0.9947 0. residual d.f. 10 10 20 10 10 20 F 944.9 718.6 599.1 1459 1870 1392 RMSE 0.139 0.161 0.157 0.121 0.110 0. t statistics in parentheses *** p<0.0001, ** p<0.001, * p<0.
Table 2 gives the weighted least squares (WLS) regression results. The weights are the numbers of C. diff. deaths in each age×sex×period cell; this is numerator-weighted rate regression, in other words. The use of death counts as weights has two appealing properties. First, it effectively down- weights data at the extremes of the age range, where deaths are fewer. This has the desirable consequence of reducing the importance of subjec- tive choices of age range (for instance, whether to use data beginning at age 40 or at age 45). Second, it moves the parameter estimates closer to those that would be achieved by maximum likelihood (ML) estimation — which can regarded as theoretically desirable (Brillinger 1986) — but with- out the computational expense of ML. To the best of our knowledge, this approach (using WLS expressly to approximate ML estimates) has not been
the level but not about the shape (slope) of the Gompertzian pattern above age 40.
To complete our empirical analysis, table 3 summarizes the under- lying causes of death when C. diff. infection was a contributory cause on the death certificate; the importance of multiple-cause analysis is noted by Dés- esquelles et al. (2014). Table 3 shows very clearly that C. diff. is involved in a wide variety of causes of death, especially those that occur in clinical set- tings. There is no single “typical” underlying cause of death when C. diff. is a contributory cause. The top 18 detailed (i.e., 4-digit) causes are listed in ta- ble 3, but these encompass only about half the deaths involving C. diff. , with the other half spread among 1,295 unique ICD codes. The clinical literature on C. diff. -associated morbidity clearly notes the association with healthcare (e.g. Sunenshine and McDonald 2006). Table 3 is fully consistent with this. For example, diseases of the circulatory system (ICD-10 Ixx.x) account for over 40% of deaths in which C. diff. is a contributory cause. While some mortality in this group may not be preceded by a hospital stay (for example, sudden death due to stroke or heart attack), admission to the emergency department and/or intensive care unit is common with these diseases (see ICD-10 “I” causes in table 3). Also, nearly 30% of deaths in which C. diff. is a contributory cause involve neoplasms (cancer, ICD-10 Cxx.x). Cancer is clearly a disease that involves interaction with healthcare prior to mortality. Moreover, immunosuppression is a risk factor for cancer (Vial and Descotes
elsewhere classified (^)
verse gut microbiome-related approaches (e.g., Rupnik 2015, Buffie et al. 2015).
One limitation of this study is that C. diff. mortality is probably the “tip of the iceberg” of infection, on which we do not have data. Moreover, the elderly, who are hospitalized disproportionately and for longer stays, may play a catalyzing role in the epidemiology of C. diff. in ways that are not captured by the mortality rates, which we have asserted are not rising in an age-specific way.
Especially because of the broad age range of increased mortality and the nonspecific nature of observed co-morbidities, public health agencies should closely monitor C. diff. mortality. The most important substantive les- son from this study is that the emergence of C. diff. as a cause of death in the United States is not due to population aging, but is healthcare-associated. We also demonstrate weighted least squares (WLS) regression as an approx- imation to full maximum likelihood for Gompertz parameter estimation, and we strongly recommend WLS over OLS whenever weights are available.
Works Cited
Bartlett, John G. 2008 a. “Historical perspectives on studies of Clostridium difficile and C. difficile infection.” Clinical Infectious Diseases 46(Suppl 1):S4– S11. ———. 2008 b. “The case for vancomycin as the preferred drug for treatment of Clostridium difficile infection.” Clinical Infectious Diseases 46(10):1489–
Bartlett, John G., Andrew B. Onderdonk, Ronald L. Cisneros, and Dennis L. Kasper. 1977. “Clindamycin-associated colitis due to a toxin-producing
species of Clostridium in hamsters.” Journal of Infectious Diseases 136(5):701–
Blossom, David B. and L. Clifford McDonald. 2007. “The challenges posed by reemerging Clostridium difficile infection.” Clinical Infectious Diseases 45(2):222–227. Brillinger, David R. 1986. “The natural variability of vital rates and associated statistics. With discussion.” Biometrics 42(4):693–734. Buffie, Charlie G., Vanni Bucci, Richard R. Stein, Peter T. McKenney, Lilan Ling, Asia Gobourne, Daniel No, Hui Liu, Melissa Kinnebrew, Agnes Viale, Eric Littmann, Marcel R. M. van den Brink, Robert R. Jenq, Ying Taur, Chris Sander, Justin R. Cross, Nora C. Toussaint, Joao B. Xavier, and Eric G. Pamer. 2015. “Precision microbiome reconstitution restores bile acid me- diated resistance to Clostridium difficile .” Nature 517(7533):205–208. Cattoir, Vincent and Roland Leclercq. 2013. “Twenty-five years of shared life with vancomycin-resistant enterococci: Is it time to divorce?” Journal of Antimicrobial Chemotherapy 68(4):731–742. Désesquelles, Aline, Elena Demuru, Viviana Egidi, Luisa Frova, France Meslé, Marilena Pappagallo, and Michele Antonio Salvatore. 2014. “Cause-specific mortality analysis: Is the underlying cause of death suffi- cient?” Quetelet Journal 2(1):119–135. Gelman, Andrew and Jonathan Auerbach. 2016. “Age-aggregation bias in mortality trends.” Proceedings of the National Academy of Sciences of the United States of America 113(7):E816–E817. Gerding, Dale N., Thomas Meyer, Christine Lee, Stuart H. Cohen, Uma K. Murthy, Andre Poirier, Trevor C. Van Schooneveld, Darrell S. Pardi, Anto- nio Ramos, Michelle A. Barron, Hongzi Chen, and Stephen Villano. 2015. “Administration of spores of nontoxigenic Clostridium difficile strain M3 for prevention of recurrent C difficile infection: A randomized clinical trial.” Journal of the American Medical Association 313(17):1719–1727. Hall, Aron J., Aaron T. Curns, L. Clifford McDonald, Umesh D. Parashar, and Ben A. Lopman. 2012. “The roles of Clostridium difficile and norovirus among gastroenteritis-associated deaths in the United States, 1999– 2007.” Clinical Infectious Diseases 55(2):216–223. Horiuchi, Shiro and Ansley J. Coale. 1982. “A simple equation for estimating the expectation of life at old ages.” Population Studies 36(2):317–326.
Pepin, Jacques. 2008. “Vancomycin for the treatment of Clostridium difficile infection: For whom is this expensive bullet really magic?” Clinical Infec- tious Diseases 46(10):1493–1498. Preston, Samuel H., Patrick Heuveline, and Michel Guillot. 2001. Demogra- phy: Measuring and modeling population processes. Blackwell, Oxford. Redelings, Matthew D., Frank Sorvillo, and Laurene Mascola. 2007. “In- crease in Clostridium difficile -related mortality rates, United States, 1999– 2004.” Emerging Infectious Diseases 13(9):1417–1419. Rupnik, Maja. 2015. “Toward a true bacteriotherapy for Clostridium difficile infection.” New England Journal of Medicine 372(16):1566–1568. Sunenshine, Rebecca H. and L. Clifford McDonald. 2006. “ Clostridium dif- ficile -associated disease: New challenges from an established pathogen.” Cleveland Clinic Journal of Medicine 73(2):187–197. Vial, Thierry and Jacques Descotes. 2003. “Immunosuppressive drugs and cancer.” Toxicology 185(3):229–240. Wachter, Kenneth W. 2014. Essential demographic methods. Harvard University Press, Cambridge. Wing, Steve and Kenneth G. Manton. 1981. “A multiple cause of death analysis of hypertension-related mortality in North Carolina, 1968–1977.” American Journal of Public Health 71(8):823–830. Wysowski, Diane K. 2006. “Increase in deaths related to enterocolitis due to Clostridium difficile in the United States, 1999–2002.” Public Health Reports 121(4):361–362. Zar, Fred A., Srinivasa R. Bakkanagari, K. M. L. S. T. Moorthi, and Melinda B. Davis. 2007. “A comparison of vancomycin and metronidazole for the treatment of Clostridium difficile -associated diarrhea, stratified by disease severity.” Clinical Infectious Diseases 45(3):302–307.
Appendix
Here we present a regression tables of the same models as in table 2, except estimated using two alternate ways from the WLS presented therein. The first is maximum likelihood (ML). Specifically, we estimate the following Poisson regression models: log (deaths) = α + βx + log (exposure) log (deaths) = α + βx + γp + δ ( p × x ) + log (exposure) ,
as an alternative to the WLS models of logged mortality rates, as in the main body of the paper; these are presented in table A-1. In table A-2, we estimate Gompertz estimates by OLS in the typical way (see, e.g., Wachter 2014, p.69; Preston et al. 2001, p.193; Horiuchi and Coale 1982). These are the exact same models as in the main text of the paper, except OLS is used instead of the weighted approach.
For both males and females, the six coefficients in the ML estima- tion in table A-1 ( α , β for each time period, and the interactive coefficients γ , δ ) are closer to the WLS (table 2) than to the OLS coefficients of table A- 2, as was asserted. The ML estimation treats each death as an observation, whereas the WLS treats each age-cell as a single data point (see “ N cells”, “ N deaths” in table A-1). This necessarily results in much higher z -statistics in table A-1, compared to the t -statistics in table 2 or table A-2; the usual cautions apply, about null hypothesis significance testing with large sample sizes. The likelihood-based estimation of coefficients, using Poisson likeli- hood (see Brillinger 1986), where each death influences the estimates, is ap- proximated by using each death as part of a weighting. The WLS approach is
Table A-2: OLS version of table 2 OLS FEMALES MALES (1) (2) (3) (1) (2) (3) Early Late Early Late (1999–2005) (2006–12) ∆ (1999–2005) (2006–12) ∆ age term ( β ) 0.127*** 0.130*** 0.130*** 0.134*** (48.6) (44.7) (47.6) (52.0) intercept ( α ) −20.17*** −19.05*** −20.31*** −19.27*** (−107) (−91.1) (−104) (−105) period term ( γ ) 1.125** 1.039** (4.01) (3.87) age×period ( δ ) 0.00294 0. (0.75) (1.06) N 12 12 24 12 12 24 R^2 0.9958 0.9950 0.9958 0.9956 0.9963 0. residual d.f. 10 10 20 10 10 20 F 2364 1997 1572 2267 2701 1791 RMSE 0.155 0.172 0.164 0.161 0.152 0. t statistics in parentheses *** p<0.0001, ** p<0.001, * p<0.