Abstract
Antimicrobial de-escalation refers to the treatment mechanism of switching from empiric antibiotics with good coverage to alternatives based on laboratory susceptibility test results, with the aim of avoiding unnecessary use of broad-spectrum antibiotics. In a previous study, we have developed multi-strain and multi-drug models in an intensive care unit setting, to evaluate the benefits and trade-offs of de-escalation in comparison with the conventional strategy called antimicrobial continuation. Our simulation results indicated that for a large portion of credible parameter combinations, de-escalation reduces the use of the empiric antibiotic but increases the probabilities of colonization and infections. In this paper, we first simplify the previous models to compare the long-term dynamical behaviors between de-escalation and continuation systems under a two-strain scenario. The analytical results coincide with our previous findings in the complex models, indicating the benefits and unintended consequences of de-escalation strategy result from the nature of this treatment mechanism, not from the complexity of the high-dimensional systems. By extending the models to three-strain scenarios, we find that de-escalation is superior than continuation in preventing outbreaks of invading strains that are resistant to empiric antibiotics. Thus decisions on antibiotic use strategies should be made specifically according to ICU conditions and intervention objectives.
Similar content being viewed by others
References
American Thoracic Society, Infectious Diseases Society of America (2005) Guidelines for the management of adults with hospital-acquired ventilator-associated, and healthcare-associated pneumonia. Am J Respir Crit Care Med 171(4):388–416. https://doi.org/10.1164/rccm.200405-644ST
Braykov NP, Morgan DJ, Schweizer ML, Uslan DZ, Kelesidis T, Weisenberg SA, Johannsson B, Young H, Cantey J, Srinivasan A et al (2014) Assessment of empirical antibiotic therapy optimisation in six hospitals: an observational cohort study. Lancet Infect Dis 14(12):1220–1227
Castillo-Chavez C, Song B (2004) Dynamical models of tuberculosis and their applications. Math Biosci Eng 1(2):361–404. https://doi.org/10.3934/mbe.2004.1.361
Cen X, Feng Z, Zheng Y, Zhao Y (2017) Bifurcation analysis and global dynamics of a mathematical model of antibiotic resistance in hospitals. J Math Biol 75(6–7):1463–1485. https://doi.org/10.1007/s00285-017-1128-3
D’Agata EMC, Webb G, Horn M (2005) A mathematical model quantifying the impact of antibiotic exposure and other interventions on the endemic prevalence of vancomycin-resistant enterococci. J Infect Dis 192(11):2004–2011. https://doi.org/10.1086/498041
D’Agata EM, Magal P, Olivier D, Ruan S, Webb GF (2007) Modeling antibiotic resistance in hospitals: the impact of minimizing treatment duration. J Theor Biol 249(3):487–499. https://doi.org/10.1016/j.jtbi.2007.08.011
De Bus L, Denys W, Catteeuw J, Gadeyne B, Vermeulen K, Boelens J, Claeys G, De Waele JJ, Decruyenaere J, Depuydt PO (2016) Impact of de-escalation of beta-lactam antibiotics on the emergence of antibiotic resistance in ICU patients: a retrospective observational study. Intensive Care Med. https://doi.org/10.1007/s00134-016-4301-z
Dellit TH, Owens RC, McGowan JE, Gerding DN, Weinstein RA, Burke JP, Huskins WC, Paterson DL, Fishman NO, Carpenter CF, Brennan PJ, Billeter M, Hooton TM (2007) Infectious diseases society of america and the society for healthcare epidemiology of america guidelines for developing an institutional program to enhance antimicrobial stewardship. Clin Infect Dis 44(2):159–177. https://doi.org/10.1086/510393
Dushoff J, Huang W, Castillo-Chavez C (1998) Backwards bifurcations and catastrophe in simple models of fatal diseases. J Math Biol 36:227–248. https://doi.org/10.1007/s002850050099
Giantsou E, Liratzopoulos N, Efraimidou E, Panopoulou M, Alepopoulou E, Kartali-Ktenidou S, Manolas K (2007) De-escalation therapy rates are significantly higher by bronchoalveolar lavage than by tracheal aspirate. Intensive Care Med 33(9):1533–1540. https://doi.org/10.1007/s00134-007-0619-x
Huang W, Cooke KL, Castillo-Chavez C (1992) Stability and bifurcation for a multiple-group model for the dynamics of HIV/AIDS transmission. SIAM J Appl Math 52(3):835–854. https://doi.org/10.1137/0152047
Hughes JS, Hurford A, Finley RL, Patrick DM, Wu J, Morris AM (2016) How to measure the impacts of antibiotic resistance and antibiotic development on empiric therapy: new composite indices. BMJ Open. https://doi.org/10.1136/bmjopen-2016-012040
Hughes J, Huo X, Falk L, Hurford A, Lan K, Coburn B, Morris A, Wu J (2017) Benefits and unintended consequences of antimicrobial de-escalation: implications for stewardship programs. PLoS ONE 12(2):e0171218
Hurford A, Morris AM, Fisman DN, Wu JH (2012) Linking antimicrobial prescribing to antimicrobial resistance in the ICU: before and after an antimicrobial stewardship program. Epidemics 4(4):203–210
Kaye KS (2012) Antimicrobial de-escalation strategies in hospitalized patients with pneumonia, intra-abdominal infections, and bacteremia. J Hosp Med 7(S1):S13–S21
Khasawneh FA, Karim A, Mahmood T, Ahmed S, Jaffri SF, Tate ME, Mehmood M (2014) Antibiotic de-escalation in bacteremic urinary tract infections: potential opportunities and effect on outcome. Infection 42(5):829–834. https://doi.org/10.1007/s15010-014-0639-8
Kim JW, Chung J, Choi SH, Jang HJ, Hong SB, Lim CM, Koh Y (2012) Early use of imipenem/cilastatin and vancomycin followed by de-escalation versus conventional antimicrobials without de-escalation for patients with hospital-acquired pneumonia in a medical ICU: a randomized clinical trial. Crit Care 16(1):R28. https://doi.org/10.1186/cc11197
Kollef MH, Morrow LE, Niederman MS, Leeper KV, Anzueto A, Benz-Scott L, Rodino FJ (2006) Clinical characteristics and treatment patterns among patients with ventilator-associated pneumonia. Chest 129(5):1210–1218
Kouyos RD, zur Wiesch PA, Bonhoeffer S (2011) Informed switching strongly decreases the prevalence of antibiotic resistance in hospital wards. PLoS Comput Biol 7(3):e1001094
Lambert ML, Suetens C, Savey A, Palomar M, Hiesmayr M, Morales I, Agodi A, Frank U, Mertens K, Schumacher M et al (2011) Clinical outcomes of health-care-associated infections and antimicrobial resistance in patients admitted to European intensive-care units: a cohort study. Lancet Infect Dis 11(1):30–38
Martcheva M (2009) A non-autonomous multi-strain SIS epidemic model. J Biol Dyn 3(2–3):235–251. https://doi.org/10.1080/17513750802638712
Martcheva M, Pilyugin SS, Holt RD (2007) Subthreshold and superthreshold coexistence of pathogen variants: the impact of host age-structure. Math Biosci 207(1):58–77. https://doi.org/10.1016/j.mbs.2006.09.010
Mermel LA, Rijnders BJA, Allon M, Craven DE, Flynn P, O’Grady NP, Raad II, Sherertz RJ, Warren DK, Bouza E (2009) Clinical practice guidelines for the diagnosis and management of intravascular catheter-related infection: 2009 update by the infectious diseases society of America. Clin Infect Dis 49(1):1–45. https://doi.org/10.1086/599376
Mokart D, Slehofer G, Lambert J, Sannini A, Chow-Chine L, Brun JP, Berger P, Duran S, Faucher M, Blache JL, Saillard C, Vey N, Leone M (2014) De-escalation of antimicrobial treatment in neutropenic patients with severe sepsis: results from an observational study. Intensive Care Med 40(1):41–49. https://doi.org/10.1007/s00134-013-3148-9
Murray TS, Egan M, Kazmierczak BI (2007) Pseudomonas aeruginosa chronic colonization in cystic fibrosis patients. Curr Opin Pediatr 19(1):83–88
Obolski U, Stein GY, Hadany L (2015) Antibiotic restriction might facilitate the emergence of multi-drug resistance. PLoS Comput Biol 11(6):1–15. https://doi.org/10.1371/journal.pcbi.1004340
Paskovaty A, Pastores SM, Gedrimaite Z, Kostelecky N, Riedel ER, Seo SK (2015) Antimicrobial de-escalation in septic cancer patients: is it safe to back down? Intensive Care Med 41(11):2022–2023. https://doi.org/10.1007/s00134-015-4016-6
Pugliese A (2000) Coexistence of macroparasites without direct interactions. Theor Popul Biol 57(2):145–165. https://doi.org/10.1006/tpbi.1999.1443
REUSSIR (2001) Le portage prolongé et la diffusion clonale interhospitalière des Pseudomonas aeruginosa multirésistants de sérotype O12 sont-ils liés? Étude multicentrique. Pathol Biol 49(8):620–623. https://doi.org/10.1016/S0369-8114(01)00219-X
Shime N, Kosaka T, Fujita N (2013) De-escalation of antimicrobial therapy for bacteraemia due to difficult-to-treat Gram-negative bacilli. Infection 41(1):203–210. https://doi.org/10.1007/s15010-012-0388-5
Spicknall IH, Foxman B, Marrs CF, Eisenberg JNS (2013) A modeling framework for the evolution and spread of antibiotic resistance: literature review and model categorization. Am J Epidemiol 178(4):508–520
Tabah A, Cotta M, Garnacho-Montero J, Roborts J, Lipman J, Tacey M, Timsit JF, Leone M, De Waele J (2016) A systematic review of the definitions, determinants, and clinical outcomes of antimicrobial de-escalation in the intensive care unit. Clin Infect Dis 62(8):1009–1017. https://doi.org/10.1093/cid/civ1199
Tepekule B, Uecker H, Derungs I, Frenoy A, Bonhoeffer S (2017) Modeling antibiotic treatment in hospitals: a systematic approach shows benefits of combination therapy over cycling, mixing, and mono-drug therapies. PLoS Comput Biol 13(9):1–22. https://doi.org/10.1371/journal.pcbi.1005745
Timsit JF, Harbarth S, Carlet J (2014) De-escalation as a potential way of reducing antibiotic use and antimicrobial resistance in ICU. Intensive Care Med 40(10):1580–1582. https://doi.org/10.1007/s00134-014-3485-3
Van Den Driessche P, Watmough J (2002) Reproduction numbers and sub-threshold endemic equilibria for compartmental models of disease transmission. Math Biosci 180:29–48. https://doi.org/10.1016/S0025-5564(02)00108-6
van Kleef E, Robotham JV, Jit M, Deeny SR, Edmunds WJ (2013) Modelling the transmission of healthcare associated infections: a systematic review. BMC Infect Dis 13:294
Webb G, Browne C (2015) A nosocomial epidemic model with infection of patients due to contaminated rooms. Math Biosci Eng 12(4):761–787. https://doi.org/10.3934/mbe.2015.12.761
Webb GF, D’Agata EMC, Magal P, Ruan S (2005) A model of antibiotic-resistant bacterial epidemics in hospitals. Proc Nat Acad Sci USA 102(37):13343–8. https://doi.org/10.1073/pnas.0504053102
Webb GF, Horn MA, D’Agata EM, Moellering RC, Ruan S (2009) Competition of hospital-acquired and community-acquired methicillin-resistant Staphylococcus aureus strains in hospitals. J Biol Dyn 48(August 2013):271. https://doi.org/10.1080/17513750903026411
Xiridou M, Borkent-Raven B, Hulshof J, Wallinga J (2009) How hepatitis D virus can hinder the control of hepatitis B virus. PLoS ONE. https://doi.org/10.1371/journal.pone.0005247
Zhao X (2003) Dynamical systems in population biology. CMS books in mathematics. Springer, Berlin
zur Wiesch PA, Kouyos R, Abel S, Viechtbauer W, Bonhoeffer S (2014) Cycling empirical antibiotic therapy in hospitals: meta-analysis and models. PLOS Pathog. https://doi.org/10.1371/journal.ppat.1004225
Acknowledgements
This research was partially supported by the University of Miami Provost’s Research Award (UM PRA 2019-409) and National Science Foundation. The author would like to thank the two anonymous reviewers and handling editor for their helpful comments and suggestions that helped improve the presentation of the paper.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
1.1 Proof of Theorem 1
Proof
Sketch of the proof: we can easily get the local stability results by analyzing the eigenvalues of the system linearized at the equilibria. The global stability results can be checked by the LaSalle’s invariant principle. When \({\mathcal {R}}_1^{\text {CT}}<1\), the Lyapunov function is defined as
hence we can get the global stability of \((S^*,X_0^*,0)\).
For \({\mathcal {R}}_1^{\text {CT}}>1\), a routine calculation gives us the unique resistant strain-prevalent equilibrium
and we can construct the Lyapunov function
to show the global stability of \(({\bar{S}}, \bar{X_0}, \bar{X_1})\). \(\square \)
1.2 Proof of Theorem 2
Proof
First, we construct a Lyapunov function to show the global stability, we continue using the notation \(X_0:=C_0+I_0\) and define \(V_3: (0,+\infty )\times (0,+\infty )\times (0,+\infty )\times [0,+\infty )\rightarrow {\mathbb {R}}\) where
Proposition 2 guarantees that the function is well-defined, taking the derivative of \(V_3\) along the solution, and we have
By the positivity of solution, we have \(\tfrac{I_0}{X_0}=\tfrac{I_0}{I_0+C_0}\le 1\). Thus, the last item in the above equation can be estimated as \(-X_1 (\mu -\beta _1 S^* - \beta _1 \tfrac{I_0}{X_0} X_0^*)\le -X_1 (\mu - \beta _1 S^* - \beta _1 X_0^*)=-X_1(\mu - \beta _1 \tfrac{\lambda }{\mu })\le 0\) for \(\beta _1< \mu ^2/\lambda \). It is also easy to check \(\mu -\beta _0 S^*\le 0\), so \(\tfrac{\mathrm{d} V_3}{\mathrm{d}t}(S, C_0, I_0, X_1)\le 0\) for \(\beta _1< \mu ^2/\lambda \).
Notice that \(\tfrac{\mathrm{d} V_3}{\mathrm{d}t}(S, C_0, I_0, X_1)= 0\) if and only if either \(\{S=S^*, X_0=X_0^*, X_1=0\}\) or \(\{S=S^*,\, X_0=X_0^*, S+I_0=\tfrac{\mu }{\beta _1}\}\). Denote the largest invariant set that contains \(\{S=S^*, X_0=X_0^*, X_1=0\}\) as \({\mathcal {M}}_1\). Then for any initial condition \((S(0),C_0(0), I_0(0), X_1(0)) \in {\mathcal {M}}_1\), we have the corresponding solution to system (4) satisfying \({\dot{S}}(0)=\dot{X_0}(0)={\dot{X}}_1(0)=0\) so there is no initial rate of change for the three compartments, and we should have \(S(t)=S^*,\,X_0(t)=X_0^*, \,X_1(t)=0\). Therefore, \({\mathcal {M}}_1\subset \{S=S^*, X_0=X_0^*, X_1=0\}\), and hence \({\mathcal {M}}_1 = \{S=S^*, X_0=X_0^*, X_1=0\}\).
Denote the largest invariant set that contains \(\{S=S^*,\, X_0=X_0^*, S+I_0=\tfrac{\mu }{\beta _1}\}\) to be \({\mathcal {M}}_2\). Then for any initial condition \((S(0),C_0(0), I_0(0), X_1(0)) \in \{S=S^*,\, X_0=X_0^*, S+I_0=\tfrac{\mu }{\beta _1}\}\), we have the corresponding solution to system (4) satisfying \({\dot{S}}(0)=\dot{X_0}(0)={\dot{X}}_1(0)=0\), then \(S(t)=S^*,\,X_0(t)=X_0^*\), and \(X_1(t)=X_1(0)\). As the population in all three compartments are constant, we have \(S^*+X_0^*+X_1(0)=\tfrac{\lambda }{\mu }\), and hence \(X_1(0)=0\); therefore, \({\mathcal {M}}_2\subset {\mathcal {M}}_1\).
For any initial condition \((S(0),C_0(0), I_0(0), X_1(0))\), the system (4) has a unique solution \((S(t), C_0(t), I_0(t), X_1(t))\) such that for any \(\varepsilon >0\), there exists a \(t_0>0\) and for all \(t>t_0\), we have \(|C_0(t)+I_0(t)-X_0^*|<\varepsilon \) and \(|X_1(t)|<\varepsilon \). And we have the estimation
thus we have
and since it is for any \(\varepsilon >0\), we know that \(\lim _{t \rightarrow +\infty } I_0(t) = I_0^*\), and hence \(\displaystyle {\lim _{t \rightarrow +\infty } C_0(t)} = C_0^*\). \(\square \)
1.3 Proof of Theorem 3
Proof
System (4) can be framed as a general system of ODEs with the bifurcation parameter \(\beta _1\):
Denote the resistant strain-free equilibrium \((S^*,C_0^*,I_0^*,0)\) as \(x^*\) and we know that
Next, we verify the assumptions of Theorem 4.1 in Castillo-Chavez and Song (2004). Firstly, the matrix of the linearized system at \(x^*\) is
and 0 is a simple eigenvalue of A and all the other eigenvalues of A are negative real numbers. Next, we compute the right eigenvector \(\mathbf{w}=(w_i)_{i=1}^{4}\) and have
where we pick
Simplifying the expressions of \(w_1\) and \(w_4\), we have
Note from the Remark 1 in Castillo-Chavez and Song (2004) \(w_1,w_2,w_3\) are not necessarily positive, but it is crucial to have \(w_4>0\) as it corresponds to a positive perturbation of the equilibrium \(x^*\) on its \(X_1\) component.
Compute the left eigenvector \(\mathbf{v}=(v_i)_{i=1}^4\) and we have \(v_1=v_2=v_3=0,\) and \(v_4=1\). We are thus able to calculate the following:
Clearly, \(w_3<0\) and if \(\beta _0<\beta _1^{\dag }\), we have \(w_1<0\), thus \(a<0\). Otherwise, if \(\beta _0\ge \beta _1^{\dag }\), we have
and thus \(a<0\), which completes the proof. \(\square \)
1.4 Proof of Proposition 4
Proof
From (14), we have
Then, condition (13) is equivalent to
which is equivalent to
Notice that
and
Thus, condition (13) is equivalent to condition (16). \(\square \)
1.5 Proof of Theorem 4
Next, we state some facts about the boundaries of region \(\Gamma \) introduced in Proposition 5.
Lemma 1
The following properties of function h are defined in (17) and \(S^*\) are easy to be verified.
- (H.1):
\(h(\beta _1)\) is a decreasing function for \(\beta _1>0\), and \(\displaystyle {\lim _{\beta _1\rightarrow 0^+} h(\beta _1)}=\tfrac{2\mu }{\beta _0}>S^*\);
- (H.2):
\(h(\beta _1)=S^*\) if and only if \(\beta _1=\beta _1^{\dag }\);
- (H.3):
\(h(\beta _1)<0\) for \(\beta _1>\tfrac{\mu ^2(\mu +\tau +\delta )}{(1-p)\lambda \delta }\);
- (H.4):
\(S^*<\tfrac{\mu +\delta }{\beta _0}\);
- (H.5):
\(\tfrac{\mu +\delta }{\beta _0}<\tfrac{\mu }{\beta _1}\) if and only if \(\beta _1<\tfrac{\mu \beta _0}{\mu +\delta }\).
With the above information, the area \(\Gamma \) is illustrated as the shaded areas in Fig. 2. We will need the following Lemma to show the existence of the resistant strain-prevalent equilibria.
Lemma 2
In the following statements, we begin to regard the function g(S) introduced in (9) with coefficients defined in (10) as a function of two variables, \(g(S,\beta _1)\). The following facts about \(g(S,\beta _1)\) are true.
- (G.1):
\(g(h(\beta _1),\beta _1)<0\) for \(\beta _1>\beta _1^{\dag }\), and \(g(h(\beta _1),\beta _1)=0\) when \(\beta _1=\beta _1^{\dag }\) and \(h(\beta _1^{\dag })=S^*\);
- (G.2):
\(g(S^*,\beta _1)=0\) if and only if \(\beta _1=\beta _1^{\dag }\) or \(\beta _1=\beta _0\), \(g(S^*,\beta _1)<0\) only when \(\beta _1\) lies in between the two roots, otherwise \(g(S^*,\beta _1)>0\);
- (G.3):
\(g(\tfrac{\mu }{\beta _1},\beta _1)>0\) for all \(\beta _1>0\);
- (G.4):
\(g(0,\beta _1)<0\) for all \(\beta _1>0\).
- (G.5):
\(g(\tfrac{\mu +\delta }{\beta _0},\beta _1)=0\) if and only if \(\beta _1=\beta _0\) or \(\beta _1=\tfrac{\mu (\mu +\delta +\tau )\beta _0}{(\mu +\delta )(\mu +\delta +\tau )-(1-p)\beta _0\lambda }\).
- (G.6):
\(\tfrac{\mu }{\beta _0}>\tfrac{\lambda }{\mu }\) implies \(\beta _1^{\dag }>\beta _0\), hence \(\beta _1^{\dag }<\beta _0\) implies \(\tfrac{\mu }{\beta _0}\le \tfrac{\lambda }{\mu }\).
- (G.7):
\(g(S,\beta _1^{\dag })=0\) has \(S^*\) as one root, and has no root in \((S^*,\min \{\tfrac{\mu +\delta }{\beta _0},\tfrac{\mu }{\beta _1^{\dag }}\})\) when \(\beta _1^{\dag }<\beta _0\).
Proof
\((G.3){-}(G.5)\) can be checked either algebraically by hand or symbolically by software Wolfram Mathematica, and we will show the proofs of the other facts.
Proof of (G.1) We first locate the solution \(\beta _1\) of \(g(h(\beta _1),\beta _1)=0\). From the proof of Proposition 4, we know that \(S=h(\beta _1)\) is a root of the following quadratic equation
$$\begin{aligned} \tfrac{\mu +\tau +\delta }{\beta _1}(S-\tfrac{\mu }{\beta _0})(S-\tfrac{\mu }{\beta _1})-(1-p)\tfrac{\lambda \delta }{\beta _0\beta _1}=0, \end{aligned}$$hence by (14), \(g(h(\beta _1),\beta _1)=f(h(\beta _1),\beta _1)(h(\beta _1)-\tfrac{\mu }{\beta _1})\). Then \(g(h(\beta _1),\beta _1)=0\) if and only if \(f(h(\beta _1),\beta _1)=0\) since \(h(\beta _1)\ne \tfrac{\mu }{\beta _1}\) for any \(\beta _1>0\). Denote the equilibrium we look for as \((h(\beta _1),I_0(\beta _1),C_0(\beta _1),X_1(\beta _1))\), from (11)–(12), we have \(f(h(\beta _1),\beta _1)=0\) if and only if \(C_0(\beta _1)=\tfrac{\lambda }{\mu }-\tfrac{\mu }{\beta _1}\) and \(X_1(\beta _1)=0\). Thus, we are looking for an equilibrium with its \(X_1\) component being 0. An easy calculation shows that we have either \(h(\beta _1)=S_+^*\) (where \(S_+^*>S^*\)) or \(h(\beta _1)=S^*\). As \(h(\beta _1)\) is a strictly decreasing function, the solution to \(h(\beta _1)=S_+^*>S^*\) is strictly less than \(\beta _1^{\dag }\). By (H.2) in Lemma 1, \(h(\beta _1)=S^*\) yields \(\beta _1=\beta _1^{\dag }\). That is, there is no \(\beta _1>\beta _1^{\dag }\) such that \(g(h(\beta _1),\beta _1)=0\).
Next we determine the sign of \(g(h(\beta _1),\beta _1)\). Observe that \(f(h(\beta _1),\beta _1)\) is a continuous function about \(\beta _1>0\), then by the intermediate value theorem, \(f(h(\beta _1),\beta _1)\) will not change its sign for \(\beta _1>\beta _1^{\dag }\). Therefore, in order to know the sign of \(f(h(\beta _1),\beta _1)\), we only need to test it at a specific \(\beta _1\) value, which we choose to be \(\tfrac{\mu ^2(\mu +\tau +\delta )}{(1-p)\lambda \delta }\) and \(h(\tfrac{\mu ^2(\mu +\tau +\delta )}{(1-p)\lambda \delta })=0\). An easy calculation yields \(f(0,\tfrac{\mu ^2(\mu +\tau +\delta )}{(1-p)\lambda \delta })=\tfrac{p\lambda }{\beta _0}(\tfrac{\delta }{\mu }+1)>0\). Also for all \(\beta _1>\beta _1^{\dag }\) we have \(h(\beta _1)<\tfrac{\mu }{\beta _1}\), hence \(g(h(\beta _1),\beta _1)=f(h(\beta _1),\beta _1)(h(\beta _1)-\tfrac{\mu }{\beta _1})<0\).
Proof of (G.2) By (10), \(g(S^*,\beta _1)\) can be seen as a quadratic function about \(\tfrac{\mu }{\beta _1}\) with positive coefficient of the second-order term, it is easy to check that \(\tfrac{\mu }{\beta _0}\) and \(\tfrac{\mu }{\beta _1^{\dag }}\) are the two roots for \(g(S^*,\beta _1)=0\), we therefore complete the proof.
Proof of (G.6) Assume \(\tfrac{\mu }{\beta _0}>\tfrac{\lambda }{\mu }\), notice that \(S^*<\min \{\tfrac{\mu }{\beta _0},\tfrac{\lambda }{\mu }\}\), then
$$\begin{aligned} \tfrac{\mu }{\beta _1^{\dag }}=\tfrac{\delta }{\mu +\tau +\delta }\tfrac{\lambda }{\mu }+\tfrac{\mu +\tau }{\mu +\tau +\delta }S^*<\tfrac{\mu }{\beta _0} \end{aligned}$$implies \(\beta _1^{\dag }>\beta _0\).
Proof of (G.7) When \(\beta _1=\beta _1^{\dag }\) we have \(\tfrac{\mu }{\beta _1^{\dag }}=\tfrac{\delta }{\mu +\tau +\delta }\tfrac{\lambda }{\mu }+\tfrac{\mu +\tau }{\mu +\tau +\delta }S^*\) and rewrite (15) as
$$\begin{aligned} g(S,\beta _1^{\dag })= & {} \left( S-\tfrac{\mu }{\beta _1^{\dag }}\right) \bigl [S^2-(\tfrac{\lambda }{\mu }-\tfrac{\tau }{\beta _0}+\tfrac{\lambda \delta }{\mu ^2}+\tfrac{\mu +\tau }{\mu }S^*)S+\tfrac{p\lambda }{\beta _0}+\tfrac{\lambda \delta }{\mu \beta _0}\bigr ]\\&+(1-p)\tfrac{\lambda \delta }{\beta _0\beta _1^{\dag }}. \end{aligned}$$Then, the roots of \(g(S,\beta _1^{\dag })=0\) can be obtained by investigating the intersections between the third-order polynomial \(y=P(S):=(S-\tfrac{\mu }{\beta _1^{\dag }})\bigl [S^2-(\tfrac{\lambda }{\mu }-\tfrac{\tau }{\beta _0}+\tfrac{\lambda \delta }{\mu ^2}+\tfrac{\mu +\tau }{\mu }S^*)S+\tfrac{p\lambda }{\beta _0}+\tfrac{\lambda \delta }{\mu \beta _0}\bigr ]\) and the horizontal line \(y=-(1-p)\tfrac{\lambda \delta }{\beta _0\beta _1^{\dag }}\).
Next we look at the roots of the quadratic equation \(S^2-(\tfrac{\lambda }{\mu }-\tfrac{\tau }{\beta _0}+\tfrac{\lambda \delta }{\mu ^2}+\tfrac{\mu +\tau }{\mu }S^*)S+\tfrac{p\lambda }{\beta _0}+\tfrac{\lambda \delta }{\mu \beta _0}\). The condition of \(\beta _1^{\dag }<\beta _0\) is equivalent to
$$\begin{aligned} \tfrac{\mu }{\beta _1^{\dag }}=\tfrac{\delta }{\mu +\tau +\delta }\tfrac{\lambda }{\mu }+\tfrac{\mu +\tau }{\mu +\tau +\delta }S^*>\tfrac{\mu }{\beta _0} \end{aligned}$$and equivalent to
$$\begin{aligned} S^*>\tfrac{\mu }{\beta _0}-\tfrac{\delta }{\mu +\tau }(\tfrac{\lambda }{\mu }-\tfrac{\mu }{\beta _0}), \end{aligned}$$then we can estimate
$$\begin{aligned} \tfrac{\lambda }{\mu }-\tfrac{\tau }{\beta _0}+\tfrac{\lambda \delta }{\mu ^2}+\tfrac{\mu +\tau }{\mu }S^*>\tfrac{\delta +\mu }{\beta _0}+\tfrac{\lambda }{\mu }. \end{aligned}$$Now, we check the determinant of the quadratic equation
$$\begin{aligned} \Delta =\left( \tfrac{\lambda }{\mu }-\tfrac{\tau }{\beta _0}+\tfrac{\lambda \delta }{\mu ^2}+\tfrac{\mu +\tau }{\mu }S^*\right) ^2 -4\tfrac{p\lambda }{\beta _0}-4\tfrac{\lambda \delta }{\mu \beta _0}>\left( \tfrac{\mu +\delta }{\beta _0}-\tfrac{\lambda }{\mu }\right) ^2. \end{aligned}$$Thus, the quadratic equation has two real roots with the bigger one
$$\begin{aligned} x_1>\max \{\tfrac{\mu +\delta }{\beta _0},\tfrac{\lambda }{\mu }\}, \end{aligned}$$it is easy to see that \(\tfrac{\mu }{\beta _1^{\dag }}=\tfrac{\delta }{\mu +\tau +\delta }\tfrac{\lambda }{\mu }+\tfrac{\mu +\tau }{\mu +\tau +\delta }S^*<\tfrac{\lambda }{\mu }\), so we have \(x_1>\tfrac{\mu }{\beta _1^{\dag }}\). Therefore, \(\tfrac{\mu }{\beta _1^{\dag }}\) is not the largest root of the third-order polynomial P(S), and the intersections between \(y=P(S)\) and line \(y=-(1-p)\tfrac{\lambda \delta }{\beta _0\beta _1^{\dag }}\) are illustrated in Fig. 8. In this way, we can see that \(S^*\) must be the smallest root of \(g(S,\beta _1^{\dag })=0\) and there is no other root lying in the interval \((S^*,\tfrac{\mu }{\beta _1^{\dag }})\), and there is no other roots lying in the even smaller interval \((S^*,\min \{\tfrac{\mu +\delta }{\beta _0},\tfrac{\mu }{\beta _1^{\dag }}\})\).
\(\square \)
Proof of Theorem 4
We only need to show the existence of an \({\widetilde{S}}\) that satisfies condition (16). Proof of 1: In the case of \(\beta _1^{\dag }>\beta _0\), for \(\beta _1>\beta _1^{\dag }>\beta _0\), we have \(\tfrac{\mu }{\beta _1}<\tfrac{\mu +\delta }{\beta _0}\); then, condition (16) is reduced to \(\max \{0,h(\beta _1)\}<{\widetilde{S}}<\tfrac{\mu }{\beta _1}\). By (G.1) and (G.4) in Lemma 2, we have \(g(\max \{0,h(\beta _1)\},\beta _1)<0\), and by (G.2) and (G.3) in Lemma2, we have \(g(\min \{S^*,\tfrac{\mu }{\beta _1}\},\beta _1)>0\). The intermediate value theorem guarantees that \(g({\widetilde{S}},\beta _1)\) has a root \({\widetilde{S}}\) in between \(\max \{0,h(\beta _1)\}\) and \(\min \{S^*,\tfrac{\mu }{\beta _1}\}\), and \({\widetilde{S}}<S^*\).
Proof of 2 We split the proof for the case of \(\beta _1^{\dag }<\beta _0\) into the following two sub-cases.
(1) If \(\tfrac{\mu }{\mu +\delta }\beta _0<\beta _1^{\dag }<\beta _0\), For \(\beta _1>\beta _1^{\dag }>\tfrac{\mu }{\mu +\delta }\beta _0\), we have \(\tfrac{\mu }{\beta _1}<\tfrac{\mu +\delta }{\beta _0}\); then, condition (16) is reduced to \(\max \{0,h(\beta _1)\}<{\widetilde{S}}<\tfrac{\mu }{\beta _1}\).
By (G.1) and (G.4) in Lemma 2, we have \(g(\max \{0,h(\beta _1)\},\beta _1)<0\), and by (G.2) and (G.3) Lemma 2, we have \(g(\min \{S^*,\tfrac{\mu }{\beta _1}\})>0\) for \(\beta _1\in (\beta _0,+\infty )\), then the intermediate value theorem guarantees that \(g({\widetilde{S}},\beta _1)\) has a root \({\widetilde{S}}\) in between \(\max \{0,h(\beta _1)\}\) and \(\min \{S^*,\tfrac{\mu }{\beta _1}\}\) for \(\beta _1\in (\beta _0,+\infty )\), and \({\widetilde{S}}<S^*\).
Again by (G.2) in Lemma 2, we have \(g(S^*,\beta _1)<0\) for \(\beta _1\in (\beta _1^{\dag },\beta _0)\), and by (G.3) in Lemma2, the intermediate value theorem guarantees that \(g({\widetilde{S}},\beta _1)\) has a root \({\widetilde{S}}\) in between \(S^*\) and \(\tfrac{\mu }{\beta _1}\) for \(\beta _1\in (\beta _1^{\dag },\beta _0)\) and clearly \({\widetilde{S}}>S^*\).
(2) If \(\beta _1^{\dag }<\tfrac{\mu }{\mu +\delta }\beta _0\), we here discuss the existence of \({\widetilde{S}}\) in three intervals of \(\beta _1\).
(2a) For \(\beta _1\in (\beta _0,+\infty )\), we have \(\tfrac{\mu }{\beta _1}<\tfrac{\mu +\delta }{\beta _0}\) and the condition (16) is reduced to \(\max \{0,h(\beta _1)\}<{\widetilde{S}}<\tfrac{\mu }{\beta _1}\). Follow the same discussion as in case (1), we have the existence of \({\widetilde{S}}<S^*\).
(2b) For \(\beta _1\in (\tfrac{\mu }{\mu +\delta }\beta _0,\beta _0)\), we have the condition (16) reduced to
By (G.2) and (G.3) in Lemma 2, we can conclude the existence of \({\widetilde{S}}>S^*\).
(2c) For \(\beta _1\in (\beta _1^{\dag },\tfrac{\mu }{\mu +\delta }\beta _0)\), we have the condition (16) reduced to
Then in order to prove the conclusion, we only need to show that \(g(\tfrac{\mu +\delta }{\beta _0},\beta _1)>0\) for \(\beta _1^{\dag }\in (\beta _1^{\dag },\tfrac{\mu }{\mu +\delta }\beta _0)\). From (10), \(g(\tfrac{\mu +\delta }{\beta _0},\beta _1)\) can be regarded as a quadratic function about \(\tfrac{\mu }{\beta _1}\) with positive coefficient of the second-order term, thus the interval for \(g(\tfrac{\mu +\delta }{\beta _0},\beta _1)>0\) is determined by the two zeros of the function which were obtained in (G.5) in Lemma 2. Firstly, if \(\tfrac{\mu (\mu +\delta +\tau )\beta _0}{(\mu +\delta )(\mu +\delta +\tau )-(1-p)\beta _0\lambda }>0\), it is easy to check that \(\tfrac{\mu }{\mu +\delta }\beta _0<\tfrac{\mu (\mu +\delta +\tau )\beta _0}{(\mu +\delta )(\mu +\delta +\tau )-(1-p)\beta _0\lambda }\), and \(g(\tfrac{\mu +\delta }{\beta _0},\beta _1)>0\) for \(\beta _1<\tfrac{\mu }{\mu +\delta }\beta _0\). Secondly, if \(\tfrac{\mu (\mu +\delta +\tau )\beta _0}{(\mu +\delta )(\mu +\delta +\tau )-(1-p)\beta _0\lambda }<0\), we have \(g(\tfrac{\mu +\delta }{\beta _0},\beta _1)>0\) if and only if \(\tfrac{\mu }{\beta _1}>\tfrac{\mu }{\beta _0}\) or \(\tfrac{\mu }{\beta _1}<\tfrac{\mu (\mu +\delta +\tau )\beta _0}{(\mu +\delta )(\mu +\delta +\tau )-(1-p)\beta _0\lambda }\) which is equivalent to \(\beta _1<\beta _0\). Thus, we have \(g(\tfrac{\mu +\delta }{\beta _0},\beta _1)>0\) for all \(\beta _1\in (\beta _1^{\dag },\tfrac{\mu }{\mu +\delta }\beta _0)\). \(\square \)
1.6 Proof of Corollary 1
Proof
It is a well-known result that if the discriminant of a third-order polynomial is strictly greater than zero, then the polynomial has three distinct real roots of multiplicity 1, hence each of the real roots S of \(g(S,\beta _1)=0\) is continuously dependent on \(\beta _1\). As (G.3) and (G.4) in Lemma 2 guarantee the existence of a real root in \((0,\tfrac{\mu }{\beta _1})\), we denote \(S^{-}(\beta _1)\) as the continuous branch that represents the smallest root in \((0,\tfrac{\mu }{\beta _1})\), and \(S^{+}(\beta _1)\) as the one that represents the largest root in \((0,\tfrac{\mu }{\beta _1})\). Then, both \(S^{-}(\beta _1)\) and \(S^{+}(\beta _1)\) are well-defined and continuous for all \(\beta _1>\beta _1^{\dag }\).
- 1.
If \(\beta _1^{\dag }>\beta _0\). For \(\beta _1\) being sufficiently large, condition (16) is equivalent to \(0<{\widetilde{S}}<\tfrac{\mu }{\beta _1}\), hence \(S^+(\beta _1)\) stays in the region
$$\begin{aligned} \Gamma _1:=\left\{ (\beta _1,S): \beta _1\ge \beta _1^{\dag },\,\max \{0,h(\beta _1)\} \le S \le \min \left\{ \tfrac{\mu }{\beta _1},S^*\right\} \right\} \end{aligned}$$for sufficiently large \(\beta _1\). By \((G.1){-}(G.4)\) in Lemma 2 and that \(S^+(\beta _1)\) is a continuous curve, \(S^+(\beta _1)\) must stay in region \(\Gamma _1\) and only touches and crosses the boundary of \(\Gamma _1\) through the point \((\beta _1^{\dag },S^*)\). Hence, both \(S^{+}(\beta _1)\) and \(S^{-}(\beta _1)\) stay below \(S^*\).
- 2.
(2a): If \(\tfrac{\mu }{\mu +\delta }\beta _0<\beta _1^{\dag }<\beta _0\). By (G.1), (G.3), (G.4), (G.7) in Lemma 2, we know that the curve \(S^{+}(\beta _1)\) stays in region
$$\begin{aligned} \Gamma _2:=\left\{ (\beta _1,S): \beta _1\ge \beta _1^{\dag },\,\max \{0,h(\beta _1)\}\le S \le \tfrac{\mu }{\beta _1}\right\} . \end{aligned}$$Then, the bifurcation branch only touches and crosses the boundary of \(\Gamma _2\) through the point \((\beta _1^{\dag },S^*)\). Also from Theorem 4, there exists a \({\widetilde{S}} \in (S^*,\tfrac{\mu }{\beta _1})\) for \(\beta _1\in (\beta _1^{\dag },\beta _0)\), so \(S^{+}(\beta _1)\) must pass through the points \((\beta _1^{\dag },S^*)\) and \((\beta _0,S^*)\) and stays above \(S^*\) for \(\beta _1\in (\beta _1^{\dag },\beta _0)\) and for \(\beta _1>\beta _0\), \(S^{+}(\beta _1)<S^*\).
(2b): If \(\beta _1^{\dag }<\tfrac{\mu }{\mu +\delta }\beta _0\). From the proof of (2c) in Theorem 4, the curve \(g(S,\beta _1)=0\) intersects the horizontal line \(S=\tfrac{\mu +\delta }{\beta _0}\) at two points and both of them are outside of the region
$$\begin{aligned} \Gamma _3:=\left\{ (\beta _1,S): \beta _1\ge \beta _1^{\dag },\,\max \{0,h(\beta _1)\}\le S \le \min \{\tfrac{\mu +\delta }{\beta _0},\tfrac{\mu }{\beta _1}\}\right\} . \end{aligned}$$By \((G.1){-}(G.4),(G.7)\) in Lemma 2, we know that the bifurcation branch \(S^{+}(\beta _1)\) stays in region \(\Gamma _3\), it only touches and crosses the boundary of \(\Gamma _3\) through the point \((\beta _1^{\dag },S^*)\), but for \(\beta _1\in (\beta _1^{\dag },\beta _0)\), we have \(S^{+}(\beta _1)>S^*\) and for \(\beta _1>\beta _0\), \(S^{+}(\beta _1)<S^*\).\(\square \)
1.7 Proof of Corollary 2
Proof
If the discriminant of a third-order polynomial is strictly smaller than zero, then the polynomial has only one real root of multiplicity 1, hence \({\widetilde{S}}\) is unique and continuously dependent on \(\beta _1\). From Theorem 4, we have that \({\widetilde{S}}<S^*\). \(\square \)
1.8 Proof of Theorem 5
We will apply Theorem 1.3.2 in Zhao (2003) to show the uniform persistence, and hence we need to formulate the problem as follows.
Consider the solution space \({\mathbb {X}}:=\{(S,I_0,C_0,X_1):S,I_0,C_0, X_1 \ge 0\},\) the interior subspace of \({\mathbb {X}}\): \({\mathbb {X}}_0:=\{(S,I_0,C_0,X_1):S,I_0,C_0, X_1 > 0\},\) and the boundary of \({\mathbb {X}}_0\) in \({\mathbb {X}}\): \(\partial {\mathbb {X}}_0:={\mathbb {X}}\setminus {\mathbb {X}}_0=\{(S,I_0,C_0,X_1):S=0 \,\text {or} \,I_0=0\,\text {or}\,C_0=0\,\text {or}\, X_1 = 0\}.\) We denote \(\Phi _t:{\mathbb {X}}\rightarrow {\mathbb {X}}\), \(t\ge 0\) as the semiflow defined by the solution of system (4), and \(M_{\partial }:=\{x\in \partial {\mathbb {X}}_0:\Phi _t(x)\in \partial {\mathbb {X}}_0,\,t\ge 0\}.\) Denote d(x, y) as the Euclidean distance between two points \(x,y\in {\mathbb {R}}^4\), and \(d(x,E):=\inf \{d(x,y):y\in E\}\) as the distance between a point x and a set \(E\subseteq {\mathbb {R}}^4\). As \({\mathbb {X}}_0\) is positively invariant, \(p(x):=d(x,\partial {\mathbb {X}}_0)\) can be seen as a generalized distance function for \(\Phi \).
Further, we denote the baseline system with only strain 0 as
Lemma 3
\(M_{\partial }=\{(S,I_0,C_0,X_1):X_1=0\}\cap \partial {\mathbb {X}}_0\).
Proof
If \(x\in \{(S,I_0,C_0,X_1):X_1=0\}\cap \partial {\mathbb {X}}_0\), then \(\Phi _t(x)\) is the solution to system (20) and \(\Phi _t(x)\in \{(S,I_0,C_0,X_1):X_1=0\}\cap \partial {\mathbb {X}}_0\subseteq M_{\partial }\); hence, \(\{(S,I_0,C_0,X_1):X_1=0\}\cap \partial {\mathbb {X}}_0\subseteq M_{\partial }.\) For any \(x\in M_{\partial }\), suppose for contradiction that \(x\in \{(S,I_0,C_0,X_1):X_1>0\}\cap \partial {\mathbb {X}}_0\); then, Proposition 2 implies \(\Phi _t(x)\in \{(S,I_0,C_0,X_1):S,I_0,C_0,X_1>0\}\not \subseteq \partial {\mathbb {X}}_0\). Therefore, we have \(x\in \{(S,I_0,C_0,X_1):X_1=0\}\cap \partial {\mathbb {X}}_0\). \(\square \)
Lemma 4
\(\{(S^*,I_0^*,C_0^*,0)\}\) is isolated in \({\mathbb {X}}\) for \(\beta _1>\beta _1^{\dag }\).
Proof
Choose \(\varepsilon =\bigl (\beta _1(S^*+I_0^*)-\mu \bigr )/3>0\), denote \({\mathcal {B}}_0\) as the ball centered at \((S^*,I_0^*,C_0^*,0)\) with radius \(\varepsilon \), and denote the maximal invariant set in \({\mathcal {B}}_0\) as M. For any point \(x\in {\mathcal {B}}_0\setminus \{(S^*,I_0^*,C_0^*,0)\}\) and \(\Phi _t(x)\in M\subseteq {\mathcal {B}}_0\) for all \(t>0\), denote \((S(t),I_0(t),C_0(t),X_1(t)):=\Phi _t(x)\), then we have \(|S(t)-S^*|<\varepsilon \) and \(|I_0(t)-I_0^*|<\varepsilon \) for all \(t>0\). However, from the last equation of system (4), we have \(\dot{X_1}(t)=[\beta _1(S(t)+I_0(t))-\mu ]X_1(t)>[\beta _1(S^*+I_0^*)-2\varepsilon _0-\mu ]X_1(t)=\varepsilon X_1(t)\) for all \(t>0\), then \(\dot{X_1}(t)>\varepsilon _1 X_1(t)\) implies \(\lim \limits _{t\rightarrow +\infty } X(t)=+\infty \) which leads to a contradiction. Therefore, the maximal invariant set in \({\mathcal {B}}_0\) is the set that contains only the equilibrium \(\{(S^*,I_0^*,C_0^*,0)\}\). \(\square \)
Proof of Theorem 5
Apply Theorem 1.3.2 in Zhao (2003) to show the uniform persistence. By Proposition 2, \({\mathbb {X}}_0\) is positively invariant for the semiflow \(\Phi _t\) generated by the solution to system (4), and \(\Phi _t\) is compact and point dissipative, hence there is a global attractor for \(\Phi _t\).
Let \(\{(S^*,I_0^*,C_0^*,0)\}\) be the finite sequence that only consists of one set, which is obviously disjoint, compact, and from Lemma 4 it is an isolated invariant set in \(\partial {\mathbb {X}}_0\). Next, we check with the following properties.
- (a):
From Lemma 3, for any \(x\in M_{\partial }\) we know that \(\Phi _t(x)\) is the solution to system (20) and by our analysis of the baseline system, we have \(\omega (x)=\{(S^*,I_0^*,C_0^*,0)\}\).
- (b):
There is apparently no subset of \(\{(S^*,I_0^*,C_0^*,0)\}\) that forms a cycle.
- (c):
\(\{(S^*,I_0^*,C_0^*,0)\}\) is isolated in \({\mathbb {X}}\) from Lemma 4.
- (d):
As stated before, we define \(p(x):=d(x,\partial {\mathbb {X}}_0)\), we need to show that for any \(x\in {\mathbb {X}}\setminus \partial {\mathbb {X}}_0\), \(\lim \limits _{t\rightarrow +\infty } d(\Phi _t(x),(S^*,I_0^*,C_0^*,0))\ne 0\), and similar argument as the proof in Lemma 4 is sufficient.
It follows from Theorem 1.3.2 in Zhao (2003) that there exists \(\eta >0\) such that for any \(x\in {\mathbb {X}}\setminus M_{\partial }\), we have \(\min \limits _{y\in \omega (x)}p(y)>\eta .\) That is,
for any nonnegative initial condition \((S(0),I_0(0),C_0(0),X_1(0))\) with \(X_1(0)>0\). \(\square \)
1.9 Proof of Theorem 6
Proof
Firstly, we compute the difference between \(\beta _1^{\dag }\) and \(\beta _0\):
Therefore, \(\beta _1^{\dag }-\beta _0\) has the same sign with the following function \(d(\beta _0)\):
Clearly, \(d(0)=\mu >0\), and
thus \(d(\beta _0)<0\) for \(\beta _0\) being sufficiently large. Since \(d(\beta _0)\) is a continuous function, there exists at least one intersection between its curve and the positive branch of the \(\beta _0\) axis. Observe that \(d(\beta _0)=0\) has at most two real roots; hence, we can conclude that there exists a unique positive solution to \(d(\beta _0)=0\), and we denote it as \(\beta _0^*\). So \(d(\beta _0)>0\) on \((0,\beta _0^*)\) and \(d(\beta _0)<0\) on \((\beta _0^*,+\infty )\), which completes the proof. \(\square \)
Rights and permissions
About this article
Cite this article
Huo, X. Modeling Antibiotic Use Strategies in Intensive Care Units: Comparing De-escalation and Continuation. Bull Math Biol 82, 6 (2020). https://doi.org/10.1007/s11538-019-00686-x
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s11538-019-00686-x