Scrub typhus, caused by the obligate intracellular bacterium Orientiatsutsugamushi, represents a significant but often overlooked cause of acute febrile illness across the Asia-Pacific region. Transmitted by larval trombiculid mites (chiggers), this zoonotic disease affects an estimated one million people annually and places over a billion people at risk worldwide [1, 2]. Despite its substantial disease burden throughout the “tsutsugamushi triangle” spanning from northern Japan to Australia and westward to Pakistan, scrub typhus remains severely under diagnosed and underreported, leading public health experts to classify it among the world’s most neglected tropical diseases [3, 4].

In India, scrub typhus has shown a concerning pattern of re-emergence over the past two decades, with significant outbreaks in previously non-endemic north-eastern states including Assam [5–7]. The clinical manifestations span a remarkably wide spectrum, ranging from mild self-limiting febrile illness to severe multi-organ dysfunction syndrome with mortality rates reaching 30–70% in untreated cases [8,9]. Severe complications frequently involve multiple organ systems, resulting in acute respiratory distress syndrome, acute kidney injury, meningoencephalitis, and other life-threatening conditions [10, 11]. Several critical knowledge gaps persist in our understanding of scrub typhus in north-eastern India. While certain risk factors for poor outcomes have been identified in other regions, these have not been systematically validated in north-eastern Indian populations. Additionally, factors specifically associated with prolonged hospitalization (as distinct from mortality) remain poorly understood, despite their considerable implications for healthcare resource utilization [12,13]. Finally, simplified clinical prediction tools for risk stratification have not been developed for this region.

This study addresses these knowledge gaps through a comprehensive analysis of scrub typhus in Southern Assam. Unlike previous work focusing predominantly on mortality, we examine both mortality and prolonged hospitalization as distinct clinical endpoints with potentially different predictors. Our objectives were to:

• Characterize the clinical and pathological spectrum of scrub typhus among hospitalized patients;
• Identify predictors of in-hospital mortality;
• Identify distinct predictors of prolonged hospitalization and develop a simplified risk stratification tool; and
• Analyze patterns of organ system involvement and their association with clinical outcomes.

By identifying distinct risk profiles for mortality versus prolonged hospitalization, our findings can guide the development of tailored management protocols for different patient subgroups. The simplified risk scores developed could enable frontline healthcare workers to rapidly identify high-risk patients requiring intensive monitoring, particularly valuable in resource-constrained settings (see Supplementary Materials S1 for additional background information).

2.1 Study Design and Setting

This was a hospital-based prospective observational study conducted over 12 months at Silchar Medical College and Hospital, a tertiary care teaching hospital in Southern Assam, India. The study enrolled consecutive adult patients with clinical features consistent with scrub typhus.

2.2 Patient Selection

Patients aged ≥18 years presenting with acute febrile illness suggestive of scrub typhus were screened for inclusion. Inclusion required serological confirmation of infection, demonstrated by positive anti-Orientiatsutsugamushi IgM antibodies using Qualitative ELISA (Inbios International USA, US-FDA and CE-IVD Approved kit with Sensitivity and Specificity of more than 98%) performed in the Sentinel Surveillance Laboratory of NOHP-PCZ (National One Health Programme for Prevention and Control of Zoonoses) in the Department of Microbiology. Key exclusion criteria included conditions that could confound the interpretation of clinical and laboratory features and cases with incomplete data (detailed criteria available in Supplementary Materials S2.1).

2.3 Data Collection

Data were prospectively collected using a standardized case report form. Clinical features recorded included fever, rash, eschar, myalgia, headache, and signs indicative of organ dysfunction. Laboratory features included complete blood count, liver and renal function tests, and other relevant measurements. Patients were followed throughout their hospital stay to track clinical course, complications, and outcomes.

2.4 Study Definitions

Organ system involvement was defined using the following criteria:

• Hepatic Dysfunction: Transaminases (AST or ALT) >80 U/L OR total serum bilirubin >1.2 mg/dL
• Renal Dysfunction: Serum creatinine >1.2 mg/dL
• Respiratory Dysfunction: Presence of ARDS OR breathing difficulty with oxygen requirement
• Central Nervous System Dysfunction: Altered sensorium, seizures, or focal neurological deficits. Altered sensorium was defined as any reduction in level of consciousness, assessed using the Glasgow Coma Scale (GCS). Patients with GCS <15 were classified as having altered sensorium. All assessments were performed by attending physicians at admission using a standardized protocol. For patients with pre-existing cognitive impairment, altered sensorium was defined as an acute change from baseline as reported by caregivers.
• Hematological Dysfunction: Platelet count <1.0 lakh/mm³
• Sepsis: Sepsis was defined according to Sepsis-3 criteria [14] as suspected infection with a Sequential Organ Failure Assessment (SOFA) score ≥2 points based on admission features.
• Multi-organ dysfunction was defined as involvement of two or more organ systems.

2.5 Outcomes

Primary outcomes assessed were in-hospital mortality and prolonged hospitalization (defined as hospital stay exceeding 7 days). Secondary outcomes included need for intensive care unit admission, development of sepsis, and organ system involvement.

2.6 Statistical Analysis

Descriptive statistics were expressed as mean ± standard deviation for continuous variables and frequencies with percentages for categorical variables. Between-group comparisons were performed using appropriate parametric or non-parametric tests. Univariate and multivariate logistic regression analyses were performed to identify independent predictors of mortality and prolonged hospitalization. Firth’s penalized likelihood method was employed to address the issue of rare outcomes [15, 16].

2.6.1 Competing Risk Analysis

To address the potential bias introduced by early mortality artificially shortening observed hospital stays, we implemented a competing risk analysis. We utilized a Fine-Gray sub distribution hazard approach wherein in-hospital death was treated as a competing event that prevents the occurrence of the primary outcome (hospital discharge). This method allows for estimation of the true hospitalization duration by accounting for patients who died before they could complete their potential full course of hospitalization.

For each organ involvement category, we calculated cumulative incidence functions (CIFs) for both discharge and death as competing outcomes. The adjusted mean hospital stay was then estimated by computing the area under the survival curve, accounting for the competing risk of death. Sub distribution hazard ratios (SHRs) were calculated to quantify the relative discharge rates between different risk groups, with values below 1.0 indicating longer hospitalization.

2.6.2 Model Development and Variable Selection

For multivariate model development, variables were initially selected based on clinical relevance and statistical significance (p<0.2) in univariate analysis. Final model variables were determined using a stepwise approach that considered statistical significance, multicollinearity, and biological plausibility. Variables with high correlation were not included simultaneously in the same model to avoid multicollinearity. The complete variable selection process, including univariate p-values and rationale for inclusion or exclusion of each candidate predictor, is detailed in Supplementary Table S14.

Timing of Variable Measurements: All predictor variables used in our risk prediction models were measured at a consistent time point to ensure appropriate temporal relationships with outcomes. Clinical features (fever, altered sensorium, breathing difficulty, jaundice) were assessed at presentation through standardized physical examination. Laboratory features including complete blood count, liver function tests, and serum creatinine were obtained from the first blood sample collected within 6 hours of admission, before any specific therapeutic interventions. Sepsis was defined according to Sepsis-3 criteria based on these admission features. For the mortality prediction model, only these baseline measurements were used as predictors to avoid incorporating disease progression markers that could introduce circular reasoning. Organ involvement was similarly determined using the admission laboratory values. This approach ensures that all predictor variables temporally precede the outcomes of interest, maintaining the prognostic rather than diagnostic nature of the developed prediction models.

2.6.3 Risk Score Development

Simplified risk scores for both mortality and prolonged hospitalization were developed based on the most significant predictors from multivariate analyses, with point values assigned proportional to the regression coefficients [18]. All statistical tests were two-sided with significance set at p<0.05. Detailed statistical methodology is available in Supplementary Materials S2.2.

2.7 Ethical Considerations

This study was approved by the Institutional Ethics Committee of Silchar Medical College and Hospital (approval number: SMC/ETHICS/M1/2024/34, dated 20.05.24). Written informed consent was obtained from all participants prior to enrollment. For patients with altered sensorium or critical illness, consent was obtained from their legally authorized representatives, with deferred consent sought from patients upon recovery when possible. All patient data were de-identified during analysis using unique study identification numbers, and confidential information was secured in password-protected databases accessible only to authorized study personnel. The study was conducted in accordance with the principles of the Declaration of Helsinki [19] and adhered to Good Clinical Practice guidelines. Patients received care treatment as per set standard regardless of study participation, and no experimental interventions were performed.

3.1 Demographic and Clinical Characteristics

A total of 80 patients with confirmed scrub typhus were included in the study. The mean age was 42.46 years (range: 16–72 years), with a male predominance (55.0%). The overall mortality rate was 7.5% (6/80 patients), and 32.5% required ICU admission (Table 1). Fever was the most common presenting symptom (96.3%), followed by myalgia (58.8%) and headache (42.5%). Eschar, a characteristic finding in scrub typhus, was present in only 25.0% of patients. Detailed clinical features stratified by age group are presented in Supplementary Table S3.1.

Table 1: Demographic Characteristics and Outcomes of Scrub Typhus Patients (n=80)

ICU = Intensive Care Unit

3.2 Laboratory Findings

Analysis of laboratory features revealed significant derangements across multiple systems (Figure 1). Hematological abnormalities were common, with a mean hemoglobin of 10.74 g/dL (range: 7.8–13.7) and leukocytosis with a mean white blood cell count of 12,615/mm³ (range: 4,200–26,080). Hepatic dysfunction was evidenced by elevated serum bilirubin (mean: 1.63 mg/dL) and markedly raised transaminases (mean AST: 122.86 U/L). Renal impairment was also common, with mean serum creatinine of 2.34 mg/dL. Laboratory features showed significant differences between patients with and without prolonged hospitalization (Supplementary Table S3.2).

3.3 Organ Involvement and Complications

Multi-organ dysfunction was a prominent feature in our cohort. Overall, 73.8% of patients had hepatic involvement, making it the most commonly affected system, followed by renal dysfunction (38.8%), central nervous system manifestations (15.0%), hematological complications (11.3%), and respiratory involvement (7.5%). Only 33.8% of patients had no organ dysfunction, while 41.3% exhibited involvement of multiple organ systems. Detailed patterns of specific organ combinations are provided in Supplementary Table S3.8.

Figure 2 illustrates the relationship between the number of organs affected and clinical outcomes. We observed a clear dose-response relationship between organ involvement and hospital stay duration, ICU admission, and mortality. The mortality rate increased from 0% in patients with 0–1 organ involvement to 33.3% with three-organ involvement. Interestingly, we observed that patients with four-organ dysfunction had shorter apparent mean stays (8.5 days) than those with three-organ involvement (10.6 days). Competing risk analysis revealed this paradox was due to earlier mortality in the four-organ group (mean time to death: 3.0 days versus 9.0 days in the three-organ group). After accounting for this competing risk, the adjusted mean stay for four-organ patients was 13.0 days, representing a competing risk effect of 4.5 days attributable to early mortality. This adjustment confirmed the expected dose-response relationship between organ involvement and hospitalization duration. Detailed results of the competing risk analysis are presented in Supplementary Tables S20–S23.

Table 2: Multi-organ Dysfunction and Clinical Outcomes

P-value for trend< 0.001

3.4 Predictors of Mortality

In multivariate analysis (Figure 3), four independent predictors of mortality were identified: sepsis (adjusted OR = 40.85, 95% CI: 5.20–320.68, p < 0.001), serum creatinine > 1.2 mg/dL (adjusted OR = 17.12, 95% CI: 2.63–111.43, p = 0.003), altered sensorium (adjusted OR = 16.61, 95% CI: 2.58–107.05, p = 0.003), and total count > 14,000/mm³ (adjusted OR = 14.44, 95% CI: 2.39–87.31, p = 0.004). Detailed univariate analyses and comparative features between survivors and non-survivors are presented in Supplementary Tables S3.4 and S3.5.

3.5 Predictors of Prolonged Hospitalization

Multivariate logistic regression identified five independent predictors of prolonged hospitalization (Figure 4): sepsis (adjusted OR = 7.03, 95% CI: 2.01–24.60, p = 0.002), altered sensorium (adjusted OR = 5.16, 95% CI: 1.19–22.35, p = 0.028), serum bilirubin >1.2 mg/dL (adjusted OR = 4.76, 95% CI: 1.50–15.13, p = 0.008), total leukocyte count >12,000/mm³ (adjusted OR = 3.94, 95% CI: 1.26–12.31, p = 0.019), and serum creatinine >1.2 mg/dL (adjusted OR = 2.89, 95% CI: 1.01–8.26, p = 0.048). Time-to-discharge analysis using Cox proportional hazards modeling confirmed these findings, with details provided in Supplementary Table S3.7.

Figure 4: Forest Plot of Multivariate Logistic Regression for Prolonged Hospitalization

Forest plot showing adjusted odds ratios with 95% confidence intervals for predictors of hospital stay >7 days. Sepsis and elevated laboratory features (serum bilirubin, leukocytosis, and creatinine) were independent predictors. The vertical dashed line represents an odds ratio of 1.0 (no effect).

3.6 Risk Stratification Tools

Based on multivariate analysis, we developed simplified risk scores for both mortality and prolonged hospitalization (Table 3). The mortality risk score incorporated sepsis (+4 points), altered sensorium (+3 points), creatinine >1.2 mg/dL (+3 points), and leukocytosis>14,000/mm³ (+2 points), with a total range of 0–12 points. Patients were classified as low risk (0–3 points, observed mortality 0%), moderate risk (4–7 points, observed mortality 17.4%), and high risk (8–12 points, observed mortality 25.0%). Characteristics of mortality cases and their risk scores are provided in Supplementary Table S3.6.

The prolonged hospitalization risk score included sepsis (+2 points), altered sensorium (+2 points), serum bilirubin >1.2 mg/dL (+1 point), total count >12,000/mm³ (+1 point), and creatinine >1.2 mg/dL (+1 point), with a total range of 0–7 points.

Table 3: Risk Score Components and Performance

Validation Metrics

• AUC (95% CI):924 (0.862–0.971) for mortality; 0.911 (0.845–0.958) for prolonged hospitalization
• Sensitivity:33% for mortality; 85.19% for prolonged hospitalization
• Specificity:19% for mortality; 86.79% for prolonged hospitalization

Low-risk patients (0–3 points) had a prolonged hospitalization rate of 15.5%, while high-risk patients (≥4 points) had a rate of 81.8%. Both models demonstrated excellent discrimination (AUC = 0.924 for mortality; AUC = 0.911 for prolonged hospitalization) and calibration (Hosmer-Lemeshow p = 0.84 and p = 0.76, respectively). Detailed model validation metrics, performance characteristics across different thresholds, and decision curve analyses are presented in Supplementary Tables S3.9 and S3.10 and Supplementary Figure S3.1.

In this hospital-based study of patients with confirmed scrub typhus from southern Assam, we observed a 7.5% mortality rate, consistent with prior estimates from India [5,12]. However, our analysis expands upon earlier findings by integrating organ system involvement patterns, laboratory markers, and clinical presentations with detailed outcome analyses, supported by rigorous statistical techniques (see Supplementary Sections S1-S3).

Our study demonstrates that neurological and respiratory manifestations, particularly altered sensorium and breathing difficulty, strongly correlate with both ICU admission and mortality. These findings reinforce results from prior studies in India [20] and East Asia [21], suggesting that neurologic involvement is a conserved pathway to poor outcomes. Importantly, we showed that altered sensorium alone carried an adjusted odds ratio (OR) of 16.61 (95% CI: 2.58-107.05) for mortality (Table S14), in alignment with global data [22, 23].

The organ system involvement analysis (Section S3 and Table S18) revealed that combinations of hepatic, renal, and CNS dysfunction were associated with significantly prolonged hospital stays and higher risk of death. These patterns were often misrepresented when analyzed without adjusting for competing risks. Notably, patients with four-organ involvement appeared to have shorter stays than those with three, a paradox resolved via Fine-Graysubdistribution hazard modeling, highlighting the impact of early mortality on observed lengths of stay (see Supplementary Discussion S4.2).

Our application of Firth's penalized logistic regression (Section S1) was crucial in adjusting for small-sample bias, especially given the limited number of mortality events (6/80). This approach produced narrower and more stable confidence intervals, offering an improvement over conventional maximum likelihood methods, particularly in rare event modeling [24].

A major contribution of this study is the development and validation of risk scores for both mortality and prolonged hospitalization. The mortality score, integrating sepsis, altered sensorium, serum creatinine, and leukocytosis, stratified patients into low, moderate, and high-risk groups with increasing mortality rates from 0% to 25% (Table S10). Similarly, the prolonged stay risk score (Table S12) effectively predicted extended hospitalizations. Both scores demonstrated strong performance in calibration (Figure S8) and discrimination (Figure S6) and offer potential for clinical deployment in endemic settings. Further external validation is needed (Supplementary Discussion S4.4).

Age-stratified analyses (Table S3 and S4) revealed increasing laboratory derangements and worse outcomes with advancing age. Older patients were more likely to exhibit multi-organ dysfunction, with elevated creatinine and bilirubin levels, aligning with age-related immune dysregulation described in previous research [25]. ICU admission and length of stay increased significantly with age, underscoring the need for intensified monitoring in elderly patients (Supplementary Section S4.3).

Our findings challenge the assumption that the presence of eschar is a consistent prognostic marker. Eschar was found in only 25% of cases and showed no significant association with outcomes (Table S1 and S6). This diverges from findings in East Asian studies, where eschar prevalence and prognostic relevance are higher [26]. The limited prognostic value in our cohort supports region-specific clinical algorithms and suggests that the absence of eschar should not delay diagnosis.

Comparison with prior Indian and global studies (Supplementary Section S4.5) confirms the regional validity of our findings while highlighting the importance of context-specific predictors. For example, hepatic involvement was more frequent (73.8%) than the global average (61.3%) [22], possibly reflecting local strain virulence or host response.

Finally, our structured approach to risk stratification can guide clinical decision-making. For instance, high-risk patients (score ≥ 8) can benefit from early ICU admission and aggressive care (Supplementary Section S4.4). We also propose re-scoring at 48-72 hours to capture clinical deterioration.

Limitations. This single-center study relied on IgM ELISA, which has a known cross-reactivity potential, especially with other Rickettsial diseases and false negativity issues [27]. Also, given the limited number of outcome events (6 deaths), the multivariate mortality model operated under a low EPV (≤2). To address the potential bias associated with such sparse data, we employed Firth's penalized likelihood method [28], which is specifically designed to reduce small-sample bias in logistic regression. Additionally, we applied bootstrap resampling (1000 iterations) to estimate optimism-adjusted performance metrics, enhancing model robustness [29]. Nonetheless, findings should be interpreted with caution and validated externally in larger cohorts.

Additionally, while our sample size was sufficient for primary analyses, larger multicenter studies are needed for generalizability and external validation of the risk models.

This study presents a detailed framework for clinical risk stratification in scrub typhus using routinely available clinical and laboratory features. By disentangling predictors of mortality from those of prolonged hospitalization, we provide actionable tools for patient triage in resource-constrained settings. The internally validated risk scores allow early identification of high-risk patients, enabling timely ICU admission, individualized care planning, and targeted discharge strategies.

Importantly, our use of robust statistical techniques such as penalized regression and competing risk analysis mitigates small-sample biases and enhances interpretability of outcomes like early mortality and hospital stay duration. The regional specificity of predictors, including the limited prognostic role of eschar and the prominence of CNS dysfunction, further emphasizes the need for context-sensitive clinical algorithms in endemic zones like northeastern India.

These findings serve both as a clinical roadmap and a research scaffold for future external validation studies, molecular investigations into host-pathogen interactions, and risk-stratified interventional trials. Ultimately, this work contributes towards more precise, evidence-based management of scrub typhus, a neglected tropical disease of global relevance.

1. Guang Xu, David H. Walker, Daniel Jupiter, Peter C. Melby, and Christine M. Arcari. A review of the global epidemiology of scrub typhus. PLoS Neglected Tropical Diseases, 11(11):e0006062, 2017.
2. Ana Bonell, YoelLubell, Paul N. Newton, John A. Crump, and Daniel H. Paris. Estimating the burden of scrub typhus: A systematic review. PLoS Neglected Tropical Diseases, 11(9):e0005838, 2017.
3. Ju Jiang and Allen L. Richards. Scrub typhus: no longer restricted to the tsutsugamushi triangle. Tropical Medicine and Infectious Disease, 7(2):14, 2022.
4. Daniel H. Paris and Nicholas P.J. Day. Tropical rickettsial infections. Clinical Medicine, 13(6):561–568, 2013.
5. George M. Varghese, Paul Trowbridge, JaishreeJanardhanan, Kuryan Thomas, John V. Peter, and Prakash Mathews. Scrub typhus in south India: clinical and laboratory manifestations, genetic variability, and outcome. International Journal of Infectious Diseases, 17(11):e981–e987, 2013.
6. K. Sharma, R. Ramakrishnan, Y.J. Hutin, A.K. Barui, P. Manickam, M. Kakkar, and M.D. Gupte. Scrub typhus in Darjeeling, India: opportunities for simple, practical prevention measures. Transactions of the Royal Society of Tropical Medicine and Hygiene, 103(11):1153–1158, 2009.
7. Vivek Kumar, Vikash Kumar, Ashok Kumar Yadav, SreenivasIyengar, Ashish Bhalla, Navneet Sharma, Ritesh Aggarwal, Sanjay Jain, and VivekanandJha. Scrub typhus is an under-recognized cause of acute febrile illness with acute kidney injury in India. PLoS Neglected Tropical Diseases, 8(1):e2605, 2014.
8. John Victor Peter, Thomas I. Sudarsan, John A.J. Prakash, and George M. Varghese. Severe scrub typhus infection: clinical features, diagnostic challenges, and management. World Journal of Critical Care Medicine, 4(3):244–250, 2015.
9. Andrew J. Taylor, Daniel H. Paris, and Paul N. Newton. A systematic review of mortality from untreated scrub typhus (Orientiatsutsugamushi). PLoS Neglected Tropical Diseases, 9(8):e0003971, 2015.
10. Vivekanandan, A. Mani, Y.S. Priya, A.P. Singh, S. Jayakumar, and S. Purty. Outbreak of scrub typhus in Pondicherry. Journal of the Association of Physicians of India, 58:24–28, 2010.
11. Merrin Griffith, John Victor Peter, G. Karthik, K. Ramakrishna, Rajalakshmi C. Kalki, George M. Varghese, AnugrahChrispal, Kishore Pichamuthu, RamyaIyyadurai, and O.C. Abraham. Profile of organ dysfunction and predictors of mortality in severe scrub typhus infection requiring intensive care admission. Indian Journal of Critical Care Medicine, 18(8):497–502, 2014.
12. Chrispal, H. Boorugu, K.G. Gopinath, J.A. Prakash, S. Chandy, O.C. Abraham, A.M. Abraham, and K. Thomas. Scrub typhus: An unrecognized threat in South India—clinical profile and predictors of mortality. Tropical Doctor, 40(3):129–133, 2010.
13. Ruey-Jun Tsay and Feng-Yee Chang. Serious complications in scrub typhus. Journal of Microbiology, Immunology and Infection, 31(4):240–244, 2016.
14. Mervyn Singer, Clifford S. Deutschman, Christopher Warren Seymour, Manu Shankar-Hari, DjillaliAnnane, Michael Bauer, RinaldoBellomo, Gordon R. Bernard, Jean-Daniel Chiche, Craig M. Coopersmith, Richard S. Hotchkiss, Mitchell M. Levy, John C. Marshall, Greg S. Martin, Steven M. Opal, Gordon D. Rubenfeld, Tom van der Poll, Jean-Louis Vincent, and Derek C. Angus. The third international consensus definitions for sepsis and septic shock (sepsis-3). JAMA, 315(8):801–810, 2016.
15. David Firth. Bias reduction of maximum likelihood estimates. Biometrika, 80(1):27–38, 1993.
16. Georg Heinze and Michael Schemper. A solution to the problem of separation in logistic regression. Statistics in Medicine, 20(16):2409–2419, 2001.
17. Jason P. Fine and Robert J. Gray. A proportional hazards model for the subdistribution of a competing risk. Journal of the American Statistical Association, 94(446):496–509, 1999.
18. Lisa M. Sullivan, Joseph M. Massaro, and Ralph B. D’Agostino. Presentation of multivariate data for clinical use: The Framingham Study risk score functions. Statistics in Medicine, 23(10):1631–1660, 2004.
19. World Medical Association. World medical association declaration of Helsinki: Ethical principles for medical research involving human subjects. JAMA, 310(20):2191–2194, 2013.
20. Nitin Rathi and AnuradhaRathi. Rickettsial diseases in central India: Proposed clinical scoring system for early detection of spotted fever. Indian Pediatrics, 48(11):867–872, 2011.
21. Dong-Min Kim, Soo Kyung Kim, JinSeok Kim, SeungBeom Lee, MyungJin Chung, Yae-Jean Kim, Nam Yong Lee, and Sang Taek Choi. Clinical and laboratory findings associated with severe scrub typhus. BMC Infectious Diseases, 17:108, 2017.
22. George Watt and Philippe Parola. Scrub typhus: pathogenesis and clinical features. Clinical Infectious Diseases, 36(4):406–410, 2003.
23. SenakaRajapakse, Chaturaka Rodrigo, and SumadhyaDeepika Fernando. Scrub typhus: pathophysiology, clinical manifestations and prognosis. Asian Pacific Journal of Tropical Medicine, 10(1):1–10, 2017.
24. Georg Heinze and Michael Schemper. A solution to the problem of monotone likelihood in Cox regression. Biometrics, 58(1):114–119, 2002.
25. David Isaacs. The impact of ageing on the immune response to infection and vaccination. British Medical Bulletin, 96(1):105–115, 2010.
26. David J. Kelly, Paul A. Fuerst, Wei-Mei Ching, and Allen L. Richards. Scrub typhus: the geographic distribution of phenotypic and genotypic variants of Orientiatsutsugamushi. Clinical Infectious Diseases, 48(Supplement 3):S203–S230, 2009.
27. Stuart D. Blacksell, AmpaiTanganuchitcharnchai, PruksaNawtaisong, PachareeKantipong, AcharaLaongnualpanich, Nicholas P.J. Day, and Daniel H. Paris. Diagnostic accuracy of the InBios scrub typhus detect enzyme-linked immunoassay for the detection of IgM antibodies in northern Thailand. Clinical and Vaccine Immunology, 23(2):148–154, 2016.
28. Georg Heinze and Michael Schemper. A comparative investigation of methods for logistic regression with separated or nearly separated data. Statistics in Medicine, 21(16):2409–2419, 2002.
29. Eric Vittinghoff and Charles E. McCulloch. Relaxing the rule of ten events per variable in logistic and Cox regression. American Journal of Epidemiology, 165(6):710–718, 2007.