Collider Bias Assessment in Colombian Indigenous Wiwa and Kogui Populations with Chronic Gastroenteric Disorder of Likely Infectious Etiology Suggests Complex Microbial Interactions Rather Than Clear Assignments of Etiological Relevance

TL;DR

This study found that multiple microbes often co-occur in indigenous populations with chronic gut issues, making it hard to pinpoint a single cause.

Contribution

The study applies collider bias analysis to assess microbial interactions in a specific indigenous population with chronic gastroenteric disorders.

Findings

Positive associations between microorganisms were much more common than negative ones.

Blastocystis spp. showed both frequent negative and positive associations with other microbes.

Complex microbial interactions were suggested rather than clear etiological assignments.

Abstract

Multiple microbial detections in stool samples of indigenous individuals suffering from chronic gastroenteric disorder of a likely infectious origin, characterized by recurring diarrhea of variable intensity, in the rural north-east of Colombia are common findings, making the assignment of etiological relevance to individual pathogens challenging. In a population of 773 indigenous people from either the tribe Wiwa or Kogui, collider bias analysis was conducted comprising 32 assessed microorganisms including 10 bacteria (Aeromonas spp., Campylobacter spp., enteroaggregative Escherichia coli (EAEC), enteropathogenic Escherichia coli (EPEC), enterotoxigenic Escherichia coli (ETEC), Salmonella spp., Shiga toxin-producing Escherichia coli (STEC), Shigella spp./enteroinvasive Escherichia coli (EIEC), Tropheryma whipplei and Yersinia spp.), 11 protozoa (Blastocystis spp., Cryptosporidium spp., Cyclospora spp., Dientamoeba fragilis, Entamoeba coli, Entamoeba bangladeshi/dispar/histolytica/moshkovskii complex, Entamoeba histolytica, Endolimax nana, Giardia duodenalis, Iodamoeba buetschlii and Pentatrichomonas hominis), 8 helminths (Ascaris spp., Enterobius vermicularis, Hymenolepis spp., Necator americanus, Schistosoma spp., Strongyloides spp., Taenia spp. and Trichuris spp.), microsporidia (Encephalocytozoon spp.) and fungal elements (microscopically observed conidia and pseudoconidia). The main results indicated that negative associations potentially pointing towards collider bias were infrequent events (n = 14), while positive associations indicating increased likelihood of co-occurrence of microorganisms quantitatively dominated (n = 88). Microorganisms showing the most frequent negative associations were EPEC (n = 6) and Blastocystis spp. (n = 3), while positive associations were most common for Trichuris spp. (n = 16), Dientamoeba fragilis (n = 15), Shigella spp./EIEC (n = 12), Ascaris spp. (n = 11) and Blastocystis spp. (n = 10). Of note, positive associations quantitively dominated for Blastocystis spp. In conclusion, collider bias assessment did not allow clear-cut assignment of etiological relevance for detected enteric microorganisms within the assessed Colombian indigenous population. Instead, the results suggested complex microbial interactions with potential summative effects. Future studies applying alternative biostatistical approaches should be considered to further delineate respective interactions.

Tables

Table 2. Summarized presentation of microorganism-specific datasets obtained from the chosen studies and included in the calculations.

Microorganism	Detections n/n (%)	Number n of Microscopy-Based Detections	Number n of Molecular Confirmed Detections	Mean Value of Recorded Cycle Threshold (Ct) Values of Real-Time PCR (±Standard Deviation), (Number of Recorded Ct Values)
Bacteria
Aeromonas spp.	5/287 (1.7%)	n.a.	5	43.2 (±1.6), (5)
Campylobacter spp.	231/708 (32.6%)	n.a.	231	30.4 (±5.5), (174)
Enteroaggregative Escherichia coli (EAEC)	177/565 (31.3%)	n.a.	177	27.8 (±5.5), (177)
Enteropathogenic Escherichia coli (EPEC)	210/565 (37.2%)	n.a.	210	27.6% (±5.5), (208)
Enterotoxigenic Escherichia coli (ETEC)	137/565 (24.2%)	n.a.	137	27.0 (±6.1), (137)
Salmonella spp.	15/714 (2.1%)	n.a.	15	31.0 (±5.7), (10)
Shiga toxin-producing Escherichia coli (STEC)	20/128 (15.6%)	n.a.	20	23.6 (±2.1), (20)
Shigella spp./enteroinvasive Escherichia coli	120/708 (16.9%)	n.a.	120	30.0 (±6.0), (98)
Tropheryma whipplei	28/389 (7.2%)	n.a.	28	30.6 (±3.0), (28)
Yersinia spp.	4/708 (0.6%)	n.a.	4	n.d. (n.a.), (0)
Protozoa
Blastocystis spp.	391/773 (50.6%)	128	263	30.5 (±4.3), (263)
Cryptosporidium spp.	17/773 (2.2%)	0	17	34.6 (±4.6), (12)
Cyclospora spp.	72/773 (9.3%)	0	72	36.0 (±5.6), (57)
Dientamoeba fragilis	75/773 (9.7%)	0	75	29.5 (7.1), (75)
Entamoeba coli	227/729 (31.1%)	227	0	n.a.
Entamoeba bangladeshi/7íspar/histolytica/moshkovskii complex	118/730 (16.2%)	118	0	n.a.
Entamoeba histolytica	56/707 (7.9%)	n.a.	56	40.0 (±5.0), (47)
Endolimax nana	216/730 (29.6%)	216	0	n.a.
Giardia duodenalis	402/773 (52.0%)	54	348	30.0 (±5.1), (218)
Iodamoeba buetschlii	55/730 (7.5%)	55	0	n.a.
Pentatrichomonas hominis	3/729 (0.4%)	3	0	n.a.
Helminths
Ascaris spp.	130/773 (16.8%)	23	107	32.0 (±4.0), (84)
Enterobius vermicularis	31/773 (4.0%)	14	17	32.0 (±3.5), (17)
Hymenolepis spp.	126/773 (16.3%)	10	116	30.0 (±4.6), (116)
Necator americanus	86/773 (11.1%)	4	82	34.0 (±3.3), (57)
Schistosoma spp.	4/730 (0.5)	0	4	35.0 (±1.4), (4)
Strongyloides spp.	39/773 (5.0%)	0	39	35.0 (±4.0), (34)
Taenia spp.	33/773 (4.3%)	0	33	30.8 (±4.1), (33)
Trichuris spp.	158/773 (20.4%)	14	144	29.4 (±3.3), (144)
Microsporidia
Encephalocytozoon spp.	7/287 (2.4%)	0	7	36.1 (±1.8), (7)
Fungi
Conidia	41/245 (16.7%)	41	0	n.a.
Pseudoconidia	27/245 (11.0%)	27	0	n.a.

Full text

1. Introduction

In remote rural areas of the Sierra Nevada de Santa Marta in the north-east of Colombia, indigenous tribes called Wiwa and Kogui live under very resource-poor conditions, e.g., without hygiene standards (sanitation, clean drinking water, electricity, etc.) known from resource-rich industrialized countries. Instead, they live closely together with their livestock and use top dressing as a fertilizer. Therefore, smear infections, food- and water-borne infections as well as soil-transmitted infections can easily occur. Consequently, infectious gastroenteritis, which is commonly spread by such modes of transmission, is among the most pressing health issues in the local population, as recently shown by our group [1].

In both Wiwa and Kogui populations, surveillance assessments including both stool microscopy and molecular diagnostic approaches have been conducted [1,2,3]. Shortly summarized, the assessed indigenous populations suffered from the medical condition of virtually constant chronic gastroenteric disorders of a likely infectious etiology with only gradual changes in intensity over time, associated with disturbed stool frequency and consistency. As this state of gastroenteric disturbance was considered “normal” in the local population, respective complaints were socially discouraged. Consequently, the indigenous only showed up for medical assessment and therapy if disease symptoms reached unbearable or critical states, e.g., in the case of bloody diarrhea. In line with this socio-culturally maintained state of chronic gastroenteric disorder in the assessed indigenous populations, high rates of enteric pathogens were detected in their stool specimens during the surveillance assessments associated with high rates of multiple detections within the same sample, as previously described [1,2,3].

As previously shown [4,5,6,7], however, multiple pathogen detections in primarily non-sterile compartments like the human gastrointestinal tract make the assignment of etiological relevance for decisions on antimicrobial therapy challenging [4,5,6,7]. The reason is that most pathogens associated with infectious gastroenteritis, with rare exceptions like mycobacteria of the tuberculosis complex, show so-called “facultative pathogenicity”. This means that they can both occur as causative agents of infectious gastroenteritis and as harmless colonizers or elements of the microbiome. The latter can occur primarily or as a residual state following previous, clinically cured, enteric infections. The associated uncertainty regarding the assignment of etiological relevance in the case of diagnostic detections resulted in discouraging the use of highly sensitive diagnostic real-time multiplex PCR panels for the routine management of diarrhea patients in the up-to-date German guideline from November 2023 [8].

Suggestions for the inclusion of quantification or semi-quantification in the estimation of etiological relevance of detected stool pathogens have been made [5,6,7] but did not lead to the definition of broadly accepted interpretation standards so far, partly due to non-consistency of respective observations. Insofar, the definition of reliable parameters for the discrimination of etiological relevance or just co-existing colonization is still an unmet diagnostic need, in particular for high-endemicity settings where multiple pathogen detections are more the standard rather than the exception [4,5,6,7].

Indeed, there is presumably more than just quantity which decides on the etiological relevance of pathogens in the human gut. Epidemiological observations of a reduction in disease severity over time in the case of repeated exposure of an initially susceptible population to gastrointestinal pathogens in resource-poor settings point towards the development of acquired semi-immunity under such circumstances [9]. As early as in 1987, it was shown that in young children living in high-endemicity settings, rapid cycling of pathogen exposition, clearance and re-expositions is associated with waning disease severity of enteric campylobacteriosis, resulting in shifts from infections to carrier states [10,11]. Unfortunately, the specific nature of the associated semi-immunity is not yet entirely understood, although it is already harnessed in preventive medical applications. Oral vaccination against typhoid fever is such an example of induced short-term protection in the human gut. As expected, especially considering the type of application, a recent meta-analysis confirmed an association between the quantity of oral antigen exposure and the strength of associated immune protection after oral typhoid fever vaccination [12]. In the case of chronic parasite infections with enteric helminths, of note, 70% of reported pathogenicity was found to be associated with only 15% of infected individuals, in line with comparably well-characterized immunological adaptation processes, as summarized in a previous review [13].

Apart from semi-immunity mediated by adaptive immune responses, the composition of the gut microbiome is also believed to mitigate an enteric pathogen’s ability to show clinically apparent virulence. Under optimum conditions, even colonization resistance towards enteric pathogens can be mediated by favorable compositions of the gut microbiome, as suggested for both laboratory animals and humans [14,15]. And even below the microbiome level, specific co-occurrence patterns of enteric microorganisms have been suggested to alter the virulence of individual microorganisms in the human gut [16,17].

In spite of the above-mentioned challenges for the assignment of etiological relevance to enteric pathogen detections in stools samples from individuals from high-endemicity settings, tools for such assignments are nevertheless desirable to guide both prioritization of public health measures as well as individual treatment decisions. Collider bias assessment is such a tool for situations in which a defined condition can be “multi-realized” due to the differential influences of various causes leading to the same effect as reported by Tönnies and colleagues [18]. Translated to the here-described situation, chronic gastroenteric disorders of a likely infectious etiology in the Colombian indigenous people, later referred to as chronic gastroenteric disorders, can be caused by a variety of different microbial pathogens. Assuming that there is a single leading pathogen in each case of gastroenteric disorder rather than summative effects of multiple microorganisms, such leading pathogens should be negatively associated with each other within the population of indigenous individuals suffering from chronic gastroenteric disorder, in line with the premises of collider bias [18].

Consequently, we performed a collider bias analysis with the meta-data from previous surveillance studies on enteric microorganisms in Colombian indigenous individuals with chronic gastroenteric disorders [1,2,3]. The aim was the identification of potential leading pathogens based on negative associations pointing towards potential collider bias situations.

2. Materials and Methods

2.1. Study Design

The assessment was designed as an explorative, hypothesis-forming, retrospective study using meta-data from previous observational and surveillance assessments on microorganisms detected in stool samples of Colombian indigenous Wiwa and Kogui populations suffering from gastroenteric disorders.

2.2. Applied Datasets

Composite datasets were created based on the meta-data of historic surveillance assessments on Colombian indigenous Wiwa and Kogui populations from the years 2014, 2018 and 2021, as previously published [1,2,3]. These assessments included diagnostic stool results based on both microscopy and real-time PCR in varying compositions. Also, residual materials from the samples had been included in test evaluation assessments, the results of which were also used for the composite datasets. For the study, no patient-related information was included in the composite datasets derived from the meta-data of the previous assessments [1,2,3]. However, as documented in the previous works [1,2,3], the female percentage of the patients varied from 35.4% to 51.4% and the mean age varied between 24.5 and 26.5 years of the age over the assessed sub-populations.

2.3. Assumptions Regarding the Dataset

For the collider bias assessment, a number of premises were considered as guaranteed, thereby accepting associated simplifications. First of all, chronic gastroenteric disorders abundant in all participating individuals [1,2,3] were considered as a unique syndromic entity just showing gradual differences in its clinical manifestation. Second, variance in the diagnostic accuracy of the applied diagnostic approaches, as defined by sensitivity and specificity over the various included surveillance assessments, was not considered, resulting in simple binary coding of the diagnostic results for each parameter (0 = negative, 1 = positive). Available cycle threshold values (Ct-values) of real-time PCR as a means of semi-quantification were recorded in the datasets, but collider bias calculation is a qualitative approach.

Each microorganism that was detected at least once in the stool samples of the assessed Wiwa and Kogui participants was included in the assessment. This resulted in the exclusion of Ancylostoma spp., Clostridioides difficile and Vibrio spp. as parameters that were included in the real-time PCR-based screening scheme but never detected in any of the assessed individuals. Recorded microorganisms were included irrespective of whether likely etiological relevance was considered or not. This strategy resulted in the inclusion of a total of 32 parameters in the analysis, comprising 10 bacterial pathotypes, species or genera (Aeromonas spp., Campylobacter spp., enteroaggregative Escherichia coli (EAEC), enteropathogenic Escherichia coli (EPEC), enterotoxigenic Escherichia coli (ETEC), Salmonella spp., Shiga toxin-producing Escherichia coli (STEC), Shigella spp./enteroinvasive Escherichia coli (EIEC, not further discriminable based on the ipaH gene detected in real-time PCR), Tropheryma whipplei and Yersinia spp.), 11 protozoa (Blastocystis spp., Cryptosporidium spp., Cyclospora spp., Dientamoeba fragilis, Entamoeba coli, Entamoeba bangladeshi/dispar/histolytica/moshkovskii complex, Entamoeba histolytica, Endolimax nana, Giardia duodenalis, Iodamoeba buetschlii and Pentatrichomonas hominis), 8 helminths (Ascaris spp., Enterobius vermicularis, Hymenolepis spp., Necator americanus, Schistosoma spp., Strongyloides spp., Taenia spp. and Trichuris spp.) and the microsporidial genus Encephalocytozoon spp. as well as conidia and pseudoconidia as fungal elements.

Considering the assessed bacteria, only real-time PCR results were available from the surveillance assessments used for the composition of the datasets [1,2,3]. For technical reasons associated with the used real-time PCR assays (eae gene used for EPEC screening occurring in STEC as well), mono-infections with STEC were technically indistinguishable from double infections with STEC and EPEC. Considering the very high overall EPEC proportions in the study population, the double infection hypothesis was favored. In detail, from a range between 197 and 210 EPEC detections depending on the true proportion of double infections, the higher assumption of 210 EPEC detections was chosen (please also see Table 2 from the results chapter).

For the above-mentioned protozoa and helminths, either microscopic detections only or, if available, real-time PCR-based detection and/or confirmation were considered as positive results. One exemption was the discrimination of Entamoeba histolytica and E. bangladeshi/dispar/histolytica/moshkovskii complex, which are microscopically indistinguishable based on the morphology of their cysts. Accordingly, the diagnosis E. histolytica was considered in case of positive E. histolytica-specific real-time PCR only. As E. dispar-specific real-time PCR was not conducted, cyst detections matching positive E. histolytica real-time PCR results might have been derived from either the abundance of E. histolytica alone or the combination of E. histolytica and other species of the E. bangladeshi/dispar/histolytica/moshkovskii complex. Considering consistently very high cycle threshold (Ct) values in E. histolytica-specific real-time PCR indicative of very low pathogen densities below the microscopic detection threshold as well as high proportions of cyst detections but only very poor matching with E. histolytica-specific real-time PCR signals, cyst-detections were generally assigned to the E. bangladeshi/dispar/histolytica/moshkovskii complex, while potential co-infection including E. histolytica was assumed in a minority of 14 instances of microscopic cyst detections combined with positive E. histolytica PCR. Considering this overall low number of events of cyst detections matching with positive E. histolytica-specific real-time PCR (please also see Table 2 from the results chapter for details), the quantitative relevance of potentially associated failure was considered as negligible. Further noteworthy, hookworm eggs were generally attributed to Necator americanus. The reason is that no trace of abundance of Ancylostoma spp. DNA was detected in any of the samples of the assessed indigenous individuals. In two individual instances, in which atypically shaped hookworm eggs had been characterized as “Unicinaria spp.-like” by microscopists in combination with negative Necator americanus-specific real-time PCR signals, these findings were excluded as uncertain from further calculations. Similarly, three atypically shaped nematode eggs characterized as “Tricocephalus spp.-like” by microscopists and thus using the outdated term for “Trichuris” were excluded as uncertain findings due to negative Trichuris spp.-specific real-time PCR results in the same samples.

As no microsporidia-specific staining techniques had been used during the surveillance assessments [1,2,3], Encephalocytozoon spp. results were only considered in cases in which specific real-time PCR had been performed. Finally, an abundance or absence of fungal elements like conidia and pseudoconidia had only been recorded for a minority of samples from the datasets and so calculations were restricted to this minority.

2.4. Statistical Assumptions

The assignment of causal relationships in observational studies is affected by the problem of causal interference. Confounding—i.e., the interference of a “confounder” with both exposure and outcome of an observational study—can at least partially be corrected by applying various statistical methods, as detailed elsewhere [19,20,21]. Correction for confounding, however, can result in so-called “collider bias”, if “colliding” of at least two factors affects the outcome [18]. Directed acyclic graphs (DAGs), i.e., simple graphical representations of causal associations, are frequently applied to identify potential sources of collider bias without a need for a deeper understanding of the underlying calculations and to easier distinguish colliding from classic confounding [22,23]. For the particular example of gastroenteric pathogens leading to the outcome gastroenteric disorder, it is visualized in the DAG presented in Table 1 how a presumed confounder “pathogen 2” can contribute to the collider “combined pathogen effect” in individuals with the outcome “gastroenteric disorder” due to exposure to “pathogen 1”.

If, however, both “pathogen 1” and “pathogen 2” are by themselves causally associated with the outcome “gastroenteric disorder”, it is assumed that the outcome will be more frequently associated with the exposure of either one or the other than with a combination of both of them. This is because more permissions need to be fulfilled to achieve the latter, more complex situation of combined occurrence. Accordingly, multivariate statistical approaches such as (longitudinal) mediation, moderation and suppression analysis, as well as their complex interplay might be considered. The latter, as well as collider bias assessment, have hardly been statistically addressed in gastrointestinal research, for which reason the assessment of collider bias by odds ratio is considered an important starting point: a negative association between “pathogen 1” and “pathogen 2” should be assumed if they indeed are elements of a colliding. For the study presented here, this negative association was assessed by performing odds ratio analysis.

Generally, individually lacking diagnostic parameters were no criteria for the exclusion of a sample from analysis. For the odds ratio calculations for the various microorganisms, only samples that had indeed been assessed for both microorganisms were included, resulting in varying total numbers of assessed samples for the various odds ratio calculations. Odds ratios with two-sided p-values and 95% confidence intervals using Fisher’s exact test with Yates’ continuity correction and the approximation of Woolf were calculated applying the software GraphPad InStat version 3.06 (GraphPad Software, Inc., San Diego, CA, USA). Significance was accepted in the case of p < 0.05. Correction for multiple testing with techniques like the Bonferroni–Holmes approach [24] was not conducted in this hypothesis-forming explorative assessment. In line with the assumptions detailed above, potential collider bias [18] was assumed in the case of negative odds ratio observed for two different microbial parameters within the assessed population of Colombian indigenous individuals with chronic gastroenteric disorders.

2.5. Ethics

An ethical clearance for this statistical assessment was not required in line with German national law, because the assessments were not based on original research data involving humans or human samples but just on meta-data from previously published studies.

3. Results

3.1. Microbial Detections Derived from the Assessed Datasets

A total of 773 completely anonymized datasets with diagnostic results obtained with different stool samples from Colombian indigenous individuals with chronic gastroenteric disorders were analyzed. Proportions of detected bacteria within the datasets ranged from 0.6% to 37.2%. In proportional order of detection, EPEC were most frequently recorded, followed by Campylobacter spp., EAEC, ETEC, Shigella spp./EIEC, STEC, Tropheryma whipplei, Salmonella spp., Aeromonas spp. and Yersinia spp. In declining order of frequency, the protozoa Giardia duodenalis, Blastocystis spp., Entamoeba coli, Endolimax nana, E. bangladeshi/dispar/histolytica/moshkovskii complex, Dientamoeba fragilis, Cyclospora spp., Entamoeba histolytica, Iodamoeba buetschlii, Cryptosporidium spp. and Pentatrichomonas hominis, as well as the helminths Trichuris spp., Ascaris spp., Hymenolepis spp., Necator americanus, Strongyloides spp., Taenia spp., Hymenolepis spp. and Schistosoma spp., were found. Only a few microsporidia were observed, and from the microscopically detected fungal elements, conidia were slightly more frequent than pseudoconidia. Details are provided in Table 2, including descriptive semi-quantification based on cycle threshold (Ct) values of real-time PCR. Shortly summarized, intermediate target DNA quantities corresponding to Ct values between 25 and 35 were recorded for most of the real-time PCR-assessed parameters. However, for Aeromonas spp., Cyclospora spp., Entamoeba histolytica and Encephalocytozoon spp., only traces of specific pathogen-DNA were observed in the stool samples, as indicated by Ct values > 35.

3.2. Proportions and Patterns of Positive and Negative Associations Observed for the Assessed Microorganisms

As visualized in Table 3, Table 4, Table 5 and Table 6 below, positive associations between assessed microorganisms, as indicated by statistically significant odds ratios of >1 (n = 88), were much more frequent than negative associations, potentially indicating collider bias events, as suggested by statistically significant odds ratios of <1 (n = 14). Details of the odds ratio calculations are presented in Appendix A (Tables A1–A470). Of note, odds ratio calculation attempts leading to non-conclusive results due to more than one zero per row or column are not shown in Appendix A.

Focusing on the few negative associations, enteroaggregative E. coli showed the most respective signals (n = 6), followed by Blastocystis spp. (n = 3) (Table 3, Table 4 and Table 6). For all other assessed microorganisms, between 0 and 2 negative associations were recorded. Even more and as detailed in Table 4, the most negative associations comprised microorganisms that are commonly considered as enteric commensals rather than pathogens.

As detailed in Table 3 and Table 5 and additionally visualized in Table 6, positive associations were much more commonly observed. Two-digit numbers of positive associations within the study population were recorded for no less than 5 assessed microorganisms, comprising Trichuris spp. (n = 16), D. fragilis (n = 15), Shigella spp./EIEC (n = 12), Ascaris spp. (n = 11) and Blastocystis spp. (n = 10) in descending order of frequency. With the exemption of Shigella spp./EIEC, all genera and species with such high numbers of recorded positive associations were protozoan or helminth parasites.

4. Discussion

This study was conducted as a hypothesis-forming explorative assessment with the aim of identifying potential etiological relevance of detected microorganisms in stool samples from Colombian indigenous individuals with chronic gastroenteric disorders of a likely infectious etiology by applying collider bias calculation. The performed analyses led to a number of results.

First of all, statistically significant odds ratios of <1, indicating negative associations potentially pointing towards collider bias within the study population, were rare events compared to observed positive associations. Even beyond quantitative rarity, the observations are difficult to interpret because microorganisms with unlikely relevance as an independent cause of infectious gastroenteritis were primarily affected. In our assessment, enteroaggregative E. coli (EAEC) was most frequently negatively associated with other recorded microorganisms. As known from previous studies in resource-limited high-endemicity settings like tropical Tanzania [7], EAEC is similarly frequent in patients with diarrhea, such as in asymptomatic individuals. And even for international travelers, in whom semi-immunity induced by frequent re-exposure to the same pathogen is quite unlikely, EAEC was not found to be significantly associated with infectious gastroenteritis as a primary pathogen in a recent study [25]. Insofar, it seems unlikely that the observed negative associations truly prove independent etiological relevance of the EAEC detections.

Blastocystis spp. was the second microorganism most frequently associated with significant odds ratios of <1 in the present assessment. Although some authors postulate a potential etiological relevance of Blastocystis spp. in human gastroenteritis depending on factors like microbiome composition and immunological situation [26,27,28,29], this opinion is not generally accepted in the scientific world. Also, Blastocystis spp. was found to be more frequently involved in positive than in negative associations, and so one might speculate that previous assumptions on hypothetic etiological relevance of this microorganism in the human gut might in fact be due to co-occurring other infections.

At least two negative associations were found for Shigella spp./EIEC detections. Even in resource-limited high-endemicity settings, several previous studies provided epidemiological evidence of the relevance of Shigella spp./EIEC as a causative agent of infectious gastroenteritis in such areas [5,6,7,16,30]. In one of those previous assessments, even a negative association with norovirus as a potential hint for collider bias was suggested [25]. In the study presented here, however, negative associations affected the protozoa Entamoeba coli and Iodamoeba buetschlii, which are usually considered as harmless colonizers indicating poor hygiene conditions rather than relevant pathogens. Also, identical quantities of negative associations were recorded for other apathogenic protozoa such as Endolimax nana and Entamoeba coli and even for detected fungal conidia, suggesting that the relevance of the observation should not be overestimated. This is even more considerable keeping in mind the high number of positive associations observed for Shigella spp./EIEC in the here-presented study as well. Insofar, the two negative associations found for Shigella spp./EIEC detections poorly confirm any independent etiological relevance of this microorganism in the assessed Colombian indigenous individuals.

Apart from Shigella spp./EIEC, positive associations were most frequently observed for enteric helminths like Trichuris spp. and Ascaris spp., as well as for protozoa of questionable etiological relevance like Blastocystis spp. and D. fragilis. The latter, in particular, is well in line with a previous modelling with samples from a smaller number of Colombian indigenous individuals suggesting mediating effects of microbial clusters containing these microorganisms on the likelihood of abundance of other enteric microorganisms, as reported by our group [31]. Such assessments, however, would be beyond the scope of the here-presented collider-bias assessment and might be considered subjects for future studies. Although the observed positive associations from this study might point towards growth-facilitating effects of one microorganism on another, alternative explanations are nevertheless not ruled out. One similarly likely alternative explanation could be shared modes of transmission due to common habits of subgroups within the assessed study population. The here-presented study, based on calculations with anonymized datasets without inclusion of patient parameters, was not designed to decide on this question.

To summarize, this study did not indicate collider bias pointing towards a likely pronounced etiological relevance of specific single microorganisms as an explanation for the chronic gastroenteric disorders shared by the assessed indigenous individuals. Instead, there were numerous hints for positive associations, suggesting mutual growth support of various microorganisms detected within the assessed stool samples, which is in line with previous observations [16,17]. Insofar, we feel justified to postulate based on our results that collider bias assessment, at least on an individual pathogen level, is a too-simple approach to conclude on complex associations like identification of causal relationships between pathogen exposure and gastroenteric disorders in individuals threatened by constant exposure in high-endemicity settings for gastroenteric pathogens. Screening for these more complex associations based on other biostatistical approaches, as recently performed [31], should be considered for future analyses.

This study has a few limitations. First of all, and as stated in the methods chapter, its assessment has a number of premises which all need to hold true if the results of the assessment shall be considered conclusive. However, even the main assumption that all study participants basically shared the same syndromic entity of chronic gastroenteric disorders with only gradual differences in symptom expression is associated with uncertainty. In fact, the restricted logistic options during the studies providing the data [1,2,3] make the definition of the disease entity vague and only syndromically describable. Second, differences in diagnostic accuracy of applied diagnostic strategies over the included surveillance assessments [1,2,3] were considered negligible for this analysis. However, this is not necessarily the case, especially in situations in which results from highly sensitive real-time PCR and investigator-dependent microscopy were merged for the analyses. Third, the analyses included considerable numbers of parameters, but not necessarily all pathogens potentially occurring in the human gut with a particular neglect of viruses and also not other factors like chronic inflammatory or functional gastrointestinal disorders. Only parameters previously assessed in the quoted surveillance studies [1,2,3] could be used for the calculations. Fourth, the retrospective design and the associated need to work with previously obtained diagnostic data did not allow a case number calculation-based assessment. Accordingly, the study was considered as a hypothesis-forming analysis only.

5. Conclusions

In spite of the limitations stated above, the study results suggested that single specific pathogens did not dominate the etiology of the medical condition of chronic gastroenteric disorder, as observed in the previously assessed Colombian indigenous populations [1,2,3]. Instead, and in line with previous observations with the same individuals [31], complex interactions seem to have played a role, potentially including whole microbial clusters and summative effects of co-occurring microorganisms. Future assessments including other biostatistical algorithms should be considered to verify or falsify this hypothesis. In particular, the terms “confounder” as well as “collider” correspond to the statistical concepts of moderation and mediation [18]. Using statistical multivariate modeling techniques associated with these concepts, associations observed in this study may be mathematically elaborated, thus paving the way for individually tailored therapeutic interventions. Such approaches, however, would be beyond the scope of the here-presented hypothesis-forming analysis and should be considered for future follow-up assessments.