Two models for outcome prediction - a comparison of logistic regression and neural networks

OBJECTIVES

Accurately predicting disease progress from a set of predictive variables is an important aspect of clinical work. For binary outcomes, the classical approach is to develop prognostic logistic regression (LR) models. Alternatively, machine learning algorithms were proposed with artificial neural networks (ANN) having become popular over the last decades. Although some studies have compared predictive accuracies of LR and ANN models, some concerns regarding their methodological quality have been voiced. Our comparison has the advantage of being based on two large independent data sets allowing for elaborate model development and independent validation.

METHODS

From the German Stroke Database, a learning data set including 1754 prospectively recruited patients with acute ischemic stroke was used. Utilizing LR and ANN, two prognostic models were developed predicting restitution of functional independence and survival after 100 days. The resulting models were applied to classify 1470 patients with acute ischemic stroke; this test data set was collected independently from the learning data. Error fractions in the test data were determined, and differences in error fractions between the algorithms were calculated with 95% confidence intervals.

RESULTS

For most prognostic models, error fractions in the test data were below 40%. There was no difference between the algorithms except for the model predicting completely versus incompletely restituted or deceased patients (difference in error fractions = 4.01% [2.10-5.96%], p = 0.0001).

CONCLUSIONS

The conscientiously applied LR remains the gold standard for prognostic modelling; however, ANN can be an alternative automated "quick and easy" multivariate analysis.

Serum luteinizing hormone, follicle-stimulating hormone and oestradiol pattern in women undergoing pituitary suppression with different gonadotrophin-releasing hormone analogue protocols for assisted reproduction

Gonadotrophin-releasing hormone analogues (GnRH-a) are used widely in controlled ovarian stimulation (COS) cycles for assisted reproduction. At present, there is great debate about the influence of exogenous hormone activity on the hypothalamus-pituitary axis following pituitary desensitization. The objective of this comparative study was to investigate the pattern of luteinizing hormone (LH), follicle-stimulating hormone (FSH) and oestradiol in women undergoing ovarian stimulation with different GnRH-a preparations. We retrospectively analysed 201 women, aged between 27 and 43 years, who were referred consecutively to our infertility clinic between January 2002 and January 2003. All women had no endocrinopathies or occult ovarian failure as assessed by day-3 hormone profile. Women were enrolled in one of the following COS protocols: depot triptorelin long protocol (n = 38), buserelin long protocol (n = 101) or buserelin short protocol (n = 62). Recombinant FSH was used to induce ovulation. Treatment was monitored by transvaginal ultrasound scan and serum measurement of FSH, LH and oestradiol. Among the women initially included, 30 had cancelled cycles due to poor ovarian response. Serum LH levels were significantly higher in the short-protocol group compared with the long-protocol groups (p < 0.001). The number of follicles, oocyte yield, number of grade-I embryos and fertilization rate were significantly lower in the short-protocol group than in the long-protocol groups. These findings showed that LH concentrations are significantly higher in women undergoing reversible medical hypophysectomy with a GnRH-a short protocol than in women treated with a long protocol. The hypothesis of an LH ceiling is confirmed.

Predicting the intracellular water compartment using artificial neural network analysis

Artificial neural networks (ANN) are used for a wide variety of data-processing applications such as predicting medical outcomes and classifying clinical data and patients. We investigated the applicability of an ANN for estimating the intracellular water compartment for a population of 104 healthy Italians ranging in age from 19 to 68 years. Anthropometric variables, bioelectric impedance analysis (BIA) variables, and reference values for intracellular water, measured using whole-body (40)K counting (ICW(K40)), were measured for all study participants. The anthropometric variables and the impedance index (height(2)/resistance) were fed to the ANN input layer, which produced as output the estimated values for intracellular water (ICW(ANN)). We also estimated intracellular water using a BIA formula for the same population (ICW(DeLorenzo)) and another for Caucasians (ICW(Gudivaka)). Errors in the estimations generated by ANN and the BIA equations were calculated as the root mean square error (RMSE). The mean (+/-SD) reference value (ICWK40) was 25.01+/-4.50 l, whereas the mean estimated value was 15.20+/-1.79 l (RMSE=11.06 l) when calculated using ICW(DeLorenzo), 18.07+/-1.14 l (RMSE=8.72 l) when using ICW(Gudivaka), and 25.01+/-2.74 l (RMSE=3.22 l) when using ICW(ANN). Based on these results, we deduce that the ANN algorithm is a more accurate predictor for reference ICW(K40) than BIA equations.

The capabilities of artificial neural networks in body composition research

When estimating in vivo body composition or combining such estimates with other results, multiple variables must be taken into account (e. g. binary attributes such as gender or continuous attributes such as most biosignals). Standard statistical models, such as logistic regression and multivariate analysis, presume well-defined distributions (e. g. normal distribution); they also presume independence among all inputs and only linear relationships, yet rarely are these requirements met in real life. As an alternative to these models, artificial neural networks can be used. In the present work, we describe the pre-processing and multivariate analysis of data using neural network techniques, providing examples from the medical field and making comparisons with classic statistical approaches. We also address the criticisms raised regarding neural network techniques and discuss their potential improvement.

Artificial neural network analysis: a novel application for predicting site-specific bone mineral density

Dual X-ray absorptiometry (DXA), which is the most commonly used method for the diagnosis and followup of human bone health, is known to produce accurate estimates of bone mineral density (BMD). However, high costs and problems with availability may prevent its use for mass screening. The objective of the present study was to estimate BMD values for healthy persons and those with conditions known to be associated with BMD, using artificial neural networks (ANN). An ANN was used to quantitatively estimate site-specific BMD values in comparison with reference values obtained by DXA (i. e. BMD(spine), BMD(pelvis), and BMD(total)). Anthropometric measurements (i. e. sex, age, weight, height, body mass index, waist-to-hip ratio, and the sum of four skinfold thicknesses) were fed to the ANN as independent input variables. The estimates based on four input variables were generated as output and were generally identical to the reference values for all studied groups. We believe the ANN is a promising approach for estimating and predicting site-specific BMD values using simple anthropometric measurements.

[Local recurrence of breast cancer. Isolated local recurrence located at the skin of the latissimus dorsi donor flap after breast reconstruction]

The reasons for local recurrence of breast cancer are currently unknown. Inadequate resection (not RO), undetected multifocal tumors, or absence of postoperative radiation are discussed. Secondary multifocal tumors as well as ipsilateral secondary tumors have been described as potential origins. Partial local recurrences can be considered a sign for general metastasis with an isolated local "reproduction" of the initial tumor. In our patient the observation of isolated local recurrences of the initial breast cancer, located at the skin of the latissimus dorsi donor flap, may be helpful in clarifying the ongoing controversial discussion. Our case study shows that the local recurrence occurred exactly at the initial location of the primary tumor independent of the local epidermal situation. We assume that activated vertical connections exist between the fascial lymphatic system and the dermal lymphatic plexus that "directed" the way to the body wall.

Myocardial damage, inflammation and thrombin inhibition in unstable coronary artery disease

AIM

Unstable coronary artery disease (CAD) is a multifactorial disease involving both thrombotic and inflammatory processes. We have assessed the time-course and the influence of thrombin inhibitors on changes in fibrinogen and C-reactive protein levels, and their relation to myocardial ischaemia in unstable CAD.

METHODS AND RESULTS

Three hundred and twenty patients were randomized to 72 h infusion with three different doses of inogatran, a direct thrombin inhibitor, or unfractionated heparin. There were no significant differences between the treatment groups in fibrinogen or C-reactive protein levels. Overall, the fibrinogen levels were significantly increased in the first 24-96 h and still elevated at 30 days. The C-reactive protein levels showed a more pronounced increase during the first 24-96 h, but then markedly decreased over 30 days. Troponin-positive compared to troponin-negative patients had higher fibrinogen and C-reactive protein levels up to 96 h, although there was an increase compared to pre-treatment levels in both groups. A high fibrinogen level (pre-treatment top tertile) was associated with an increased rate of death or myocardial (re-)infarction at 30 days, 13% vs 5.6%, P=0.03, and increased long-term mortality. A high C-reactive protein level was related to increased 30-day mortality, 4% vs 0%, P=0.01.

CONCLUSION

Myocardial cell injury was related to a high degree of inflammation, only some of which is an acutephase response due to tissue damage. The rise in fibrinogen was sustained, which might reflect low grade inflammation with long-term risk of thrombosis. The transient elevation of C-reactive protein levels might indicate a propensity to a pronounced inflammatory response and is associated with increased mortality.

Predicting Type 2 diabetes using an electronic nose-based artificial neural network analysis

Diabetes is a major health problem in both industrial and developing countries, and its incidence is rising. Although detection of diabetes is improving, about half of the patients with Type 2 diabetes are undiagnosed and the delay from disease onset to diagnosis may exceed 10 yr. Thus, earlier detection of Type 2 diabetes and treatment of hyperglycaemia and related metabolic abnormalities is of vital importance. The objectives of the present study were to examine urine samples from Type 2 diabetic patients and healthy volunteers using the electronic nose technology and to evaluate possible application of data classification methods such as self-learning artificial neural networks (ANN) and logistic regression (LR) in comparison with principal components analysis (PCA). Urine samples from Type 2 diabetic patients and healthy controls were processed randomly using a simple 8-sensors electronic nose and individual electronic nose patterns were qualitatively classified using the "Approximation and Classification of Medical Data" (ACMD) network based on 2 output neurons, binary LR analysis and PCA. Distinct classes were found for Type 2 diabetic subjects and controls using PCA, which had a 96.0% successful classification percentage mean while qualitative ANN analysis and LR analysis had successful classification percentages of 92.0% and 88.0%, respectively. Therefore, the ACMD network is suitable for classifying medical and clinical data.

Generation of classification criteria for chronic fatigue syndrome using an artificial neural network and traditional criteria set

OBJECTIVE

The definition of chronic fatigue syndrome (CFS) is still disputed and no validated classification criteria have been published. Artificial neural networks (ANN) are computer-based models that can help to evaluate complex correlations. We examined the utility of ANN and other conventional methods in generating classification criteria for CFS compared to other diseases with prominent fatigue, systemic lupus erythematosus (SLE) and fibromyalgia syndrome (FMA).

PATIENTS AND METHODS

Ninety-nine case patients with CFS, 41 patients with SLE and 58 with FMA were recruited from a generalist outpatient population. Clinical symptoms were documented with help of a predefined questionnaire. The patients were randomly divided into two groups. One group (n = 158) served to derive classification criteria sets by two-fold cross-validation, using a) unweighted application of criteria, b) regression coefficients, c) regression tree analysis, and d) artificial neural networks in parallel. These criteria were validated with the second group (n = 40).

RESULTS

Classification criteria developed by ANN were found to have a sensitivity of 95% and a specificity of 85%. ANN achieved a higher accuracy than any of the other methods.

CONCLUSION

We present validated criteria for the classification of CFS versus SLE and FMA, comparing different classification approaches. The most accurate criteria were derived with the help of ANN. We therefore recommend the use of ANN for the classification of syndromes with complex interrelated symptoms like CFS.

Improved differentiation between Churg-Strauss syndrome and Wegener's granulomatosis by an artificial neural network

OBJECTIVE

To examine the operating characteristics of the American College of Rheumatology (ACR) classification criteria for Churg-Strauss syndrome (CSS) and Wegener's granulomatosis (WG), and to develop and validate improved criteria for distinguishing CSS from WG.

METHODS

The ACR classification criteria for WG and CSS were applied to 40 consecutive CSS patients age- and sex-matched with 40 patients with WG. Forty-three clinical, laboratory, and biopsy parameters were assessed. Artificial neural networks (ANNs) were trained and tested with all 43 parameters (set A) and with 15 solely clinical parameters documented at the initial manifestation of the disease (set B). The ANNs were trained with data from the first 27 CSS and 27 WG patients and validated with data from the next 13 consecutive CSS and 13 WG patients. To compare the ANNs with established methods, traditional format and classification tree criteria were generated using the same data sets.

RESULTS

Fourteen of 40 CSS patients fulfilled the ACR criteria for WG, while 4 WG patients met the ACR criteria for CSS. The ANN, in contrast, reliably distinguished all CSS cases from WG cases (parameter set A, accuracy 100%). For parameter set B, the ANN achieved an accuracy of 100% in the training phase and 96% for validation. The newly formulated traditional format and classification tree criteria reached an accuracy of 81% and 88%, respectively.

CONCLUSION

The ACR criteria for WG do not reliably differentiate between CSS and WG (specificity 65%). An ANN, however, could be trained to correctly allocate all but 1 patient on the basis of clinical data. Indeed, the ANN applied in this study proved superior to established methods of classification. We suggest that an ANN may be effectively applied in the classification of systemic vasculitides.

Coagulation activity and clinical outcome in unstable coronary artery disease

In the current study, we investigated molecular markers of coagulation activity, ie, prothrombin fragment 1+2 (F1+2), thrombin-antithrombin (TAT) complex, soluble fibrin (SF), and D-dimer, and their relation to death, myocardial infarction, and refractory angina during and after anticoagulant treatment in unstable coronary artery disease. Patients with unstable coronary artery disease (N=320) were randomized to a 72-hour infusion with either inogatran, a low-molecular-mass direct thrombin inhibitor, or unfractionated heparin. During the 30-day follow-up, a 40% lower event rate was seen in patients with high compared with low baseline levels of TAT or SF. High baseline levels of coagulation activity were correlated with a larger decrease during treatment. Patients with decreased compared with raised F1+2 or TAT levels after 6 hours of treatment had a 50% lower event rate at 30 days (F1+2, P=0.04; TAT, P=0.02). At the cessation of antithrombin treatment, there was a clustering of cardiac events that tended to be related to a rise in the levels of TAT and the other markers. During long-term follow-up (median, 29 months), there was a relation between higher baseline levels of D-dimer (P=0.003) and increased mortality. High baseline levels of molecular markers of coagulation activity might identify patients with a thrombotic condition (as the major cause of instability) who are good responders to anticoagulant therapy, with a larger decrease in coagulation activity during treatment and a decreased risk of ischemic events. However, this early benefit is lost during long-term follow-up when high baseline levels of coagulation activity are associated with a raised risk of early reactivation and increased mortality.

Cyclin-dependent kinase-inhibitor 1 (CDKN1A) in the squamous epithelium of the oropharynx: possible implications of molecular biology and compartmentation

The cdknlA gene encodes CDKN1A, a protein that regulates cell cycle progression, terminal differentiation, and apoptosis. Polymorphisms or loss of heterozygosity of this usually biallelically expressed gene have no major impact on carcinogenesis. The prevalence of somatic mutations in malignancies is low. Gene rearrangements involving cdknlA are scarce. CDKN1A is expressed in both premalignant and malignant lesions. While the prognostic value of nuclear CDKN1A expression is controversial, the prognostic value of its recently discovered cytoplasmic accumulation is simply unknown. CDKN1A translocates from the nucleus to the cytoplasm when cleaved by caspase-like activities during early apoptosis. The presence of cytoplasmic catabolites (e.g.: p14) might therefore indicate apoptosis. We found no correlation between nuclear and cytoplasmic anti-CDKN1A immunoreactivity in our samples of oropharyngeal squamous cell carcinoma. CDKN1A Cap20, CDKN1, CDKN1A, CDKNA1, Cip-1, Mda-6, P21, Pic1, Sdi-1, Waf-1.

Artificial neural networks for classifying olfactory signals

For practical applications, artificial neural networks have to meet several requirements: Mainly they should learn quick, classify accurate and behave robust. Programs should be user-friendly and should not need the presence of an expert for fine tuning diverse learning parameters. The present paper demonstrates an approach using an oversized network topology, adaptive propagation (APROP), a modified error function, and averaging outputs of four networks described for the first time. As an example, signals from different semiconductor gas sensors of an electronic nose were classified. The electronic nose smelt different types of edible oil with extremely different a-priori-probabilities. The fully-specified neural network classifier fulfilled the above mentioned demands. The new approach will be helpful not only for classifying olfactory signals automatically but also in many other fields in medicine, e.g. in data mining from medical databases.

Activated partial thromboplastin time and clinical outcome after thrombin inhibition in unstable coronary artery disease

Aims Direct thrombin inhibitors have failed to prove superiority over unfractionated heparin in several clinical trials of unstable coronary artery disease. We have investigated the relationship between activated partial thromboplastin time levels and adverse clinical events, i.e. death, myocardial (re-)infarction or refractory angina. Methods and Results One thousand two hundred and nine patients with unstable coronary artery disease were randomized to 72 h infusion with inogatran, a low molecular mass direct thrombin inhibitor, or unfractionated heparin. During 30 days follow-up there was no significant difference between inogatran and unfractionated heparin treatment as regards clinical outcome. 11.6% of the 464 inogatran treated patients with activated partial thromboplastin time above the median at 6 h (44 s) had a clinical event in 7 days, and 6.6% of the 423 patients with activated partial thromboplastin time below the median (P=0.01). After 30 days the event rate was still 41% higher in the inogatran patients with activated partial thromboplastin time above the median (P=0.06). Activated partial thromboplastin time in quartiles indicated a direct relationship between higher activated partial thromboplastin time and worse outcome. In contrast, during heparin infusion there was a trend for improved clinical outcome with activated partial thromboplastin time above the median, but this benefit was lost after cessation of treatment.

CONCLUSIONS

Higher activated partial thromboplastin time levels during inogatran treatment are related to increased risk of death, myocardial infarction or refractory angina. This might, at least in part, be explained by differences in anticoagulant mechanisms between direct thrombin inhibitors and heparin, and further emphasizes the poorly defined optimal activated partial thromboplastin time range during treatment with direct thrombin inhibitors in unstable coronary artery disease.

Thrombin inhibitors suppress the thrombin-thrombomodulin-mediated generation of activated protein C

In the treatment of unstable coronary artery disease, direct thrombin inhibitors have shown no or only limited benefit as compared with heparin, despite theoretical advantages. One explanation may be that the direct thrombin inhibitors to a greater extent than heparin have an inhibiting effect on the generation and activity of activated protein C. In the present study, this hypothesis was tested in an in vitro, "purified" system, where human protein C underwent activation to activated protein C by the thrombin-thrombomodulin complex. Direct thrombin inhibitors, inogatran and hirudin, unfractionated heparin+antithrombin, or dalteparin+antithrombin, were added to the system before activation to evaluate their inhibitory effect on the generation of activated protein C. At inhibitor concentrations well below the achieved plasma levels in major clinical trials, the thrombin-thrombomodulin-mediated activation of protein C was inhibited by all the studied inhibitors in a dose-dependent manner, but, contrary to our hypothesis, to a greater extent by unfractionated heparin+antithrombin and dalteparin+antithrombin than by the direct thrombin inhibitors, hirudin and inogatran. Despite difficulties to draw conclusions for the in vivo situation, the in vitro inhibition, by all studied inhibitors, of the generation of activated protein C, found in this study may be a possible explanation for ongoing cardiovascular events despite adequate treatment with thrombin inhibitors, in patients with unstable coronary artery disease. This inhibition of the generation of activated protein C may also contribute to the rebound in cardiovascular events after withdrawal of effective antithrombotic treatment.

The effect of a low molecular mass thrombin inhibitor, inogatran, and heparin on thrombin generation and fibrin turnover in patients with unstable coronary artery disease

AIM

This study evaluated a novel specific thrombin inhibitor, inogatran, in comparison with unfractionated heparin, with regard to markers for coagulation activity in patients with unstable coronary artery disease.

METHODS AND RESULTS

In the Thrombin Inhibition In Myocardial Ischaemia (TRIM) study patients were randomized to one of three different doses of inogatran or to unfractionated heparin, given intravenously over 72 h. In a subpopulation of 320 patients, markers for coagulation activity were measured at baseline, during and after the study infusion. Prothrombin fragment 1 + 2, indicating thrombin generation, decreased in the low, medium and high dose inogatran groups and in the heparin group during the first 6 h of treatment by 12%, 15%, 21% and 26%, respectively. From 6 h to 72 h after the start of infusion the levels changed by -7%, -6%, -4% and +34%, respectively. The increase in the heparin group continued after the infusion was stopped. Thrombin-antithrombin complex, also indicating thrombin generation, decreased by 0%, 2%, 18% and 22%, respectively, during the first 6 h of treatment. During the same period soluble fibrin, an intermediate in fibrin formation, increased both in the low and medium inogatran group by 9%, while a decrease by 4% and 18%, respectively, was seen in the high dose inogatran group and in the heparin group. Fibrin dissolution, as measured by fibrin D-dimer, decreased during the first 24 h of treatment by 20%, 18%, 18% and 20%, respectively. The first 24 h after discontinuation of infusion, fibrin D-dimer increased by 6%, 23%, 25% and 44%, respectively. After 72 h, at the end of infusion, patients treated with inogatran, to a larger extent than those given heparin, had suffered from death, myocardial infarction or refractory angina pectoris. After 7 days this trend was less marked.

CONCLUSION

The more pronounced decrease in thrombin generation and fibrin turnover during the first 6 h of infusion, and the later increase in thrombin generation and fibrin turnover, in the heparin group, as compared to the inogatran groups, may be related to the lower clinical event rate during infusion with heparin compared with inogatran and the recurrence of ischaemic events, early after cessation of heparin infusion.

[Effect of dressing materials on wound healing]

In the early 1970s, a major revolution in the management of wounds began. Simple woven absorbents that kept the wound as dry as possible were superseded by occlusive dressings that provided moist wound healing. The wound healing process is influenced by control of moisture content, thermal properties, gaseous permeability, pH effects, and impermeability to micro-organisms. Low adherence of the dressing provides protection of the newly formed tissue and relieves the patient's pain during removal.

Human pathogenic virus-associated pseudolymphomas and lymphomas with primary cutaneous manifestation in humans and animals

The etiologic role of viruses in cutaneous lymphoproliferative disorders is still controversial. In benign cutaneous pseudolymphomas of the human skin, human T-cell leukemia/lymphoma virus (HTLV) type I (HTLV-I), varicella zoster virus, Epstein-Barr virus (EBV), and human herpesvirus (HHV) 6 (HHV-6) are the viruses most often identified, whereas in malignant lymphoproliferation human immunodeficiency virus type 1 (HIV-1), HTLV-I/II, and EBV are more common. Coinfections with more than one virus species have occurred in a number of cases. HHV-8 in association with a lymphoproliferative lesion appears to be indicative of a malignant cutaneous lymphoma rather than of pseudolymphoma. Negative results are of no diagnostic value because of the relatively low number of virus-positive cases: a considerable proportion of studies (with a large number of subjects) have documented virus-negative findings. Perhaps with the exception of HIV-1, findings of viral infections seem to indicate secondary rather than primary infections. Reports on animal models associated with human pathogenic viruses are scarce.

[Value of preoperative diagnosis in struma surgery]

To determine the usefulness of so-called routine diagnostic procedures prior to thyroid operations, in 2007 consecutive patients undergoing thyroid surgery from 1976 until 1991, the results of 1833 scintigraphy examinations, 1358 tracheal X-rays, 773 ultrasound examinations, 302 fine-needle aspiration biopsies (FNAB), and 73 frozen sections (FS) performed during operation were evaluated. Scintigraphy and ultrasound showed similar positive predictive values, ranging between 85% and 95%, in detecting thyroid nodules. The sensitivity of FNAB and FS were 44% and 83%, respectively. Mechanical alteration of the trachea was found to correlate to a high degree with severe postoperative respiratory disorders, requiring prolonged ventilation or tracheotomy either during primary surgery or as a secondary procedure.

Oxidation of macrophage membrane cholesterol by intracellular Rhodococcus equi

Phagocytic uptake by cultured mouse macrophages (PD388D1) of a virulent strain (ATCC 33701) of Rhodococcus equi producing substantial cholesterol oxidase was accompanied by intracellular survival of the bacteria, and enzymatic oxidation of macrophage membrane cholesterol. A non-virulent strain (4219) lacking cholesterol oxidase was largely eliminated from the macrophages and did not bring about oxidation of membrane cholesterol. When R. equi 33701 was co-phagocytosed with Corynebacterium pseudotuberculosis there was a significant enhancement (10-fold) in the amount of oxidation product (4-cholesten-3-one) generated. R. equi and C. pseudotuberculosis are cooperative partners in the hemolysis of sheep erythrocytes, traceable to the cholesterol oxidase of the former, and phospholipase D of the latter. Results are discussed relative to the role of cooperative cytotoxins in damage to host tissue by bacterial pathogens.

Rhodococcus equi and Arcanobacterium haemolyticum: two "coryneform" bacteria increasingly recognized as agents of human infection

Rhodococcus equi and Arcanobacterium haemolyticum, formerly classified in the genus Corynebacterium, are members of the loosely defined taxon "coryneform" bacteria. Although they are the etiologic agents of distinct human infections, both organisms are frequently overlooked, which results in missed or delayed diagnoses. R. equi, long known as an important pathogen of immature horses, has become in the past three decades an opportunistic pathogen of severely immunosuppressed humans. Most cases are secondary to HIV infection. When specifically sought in throat swab cultures, A. haemolyticum is found responsible for 0.5% to 2.5% of bacterial pharyngitis, especially among adolescents. These two microorganisms represent a spectrum of disease in humans: from a mild, common illness to a rare life-threatening infection. Each organism elaborates lipid hydrolyzing enzymes (cholesterol oxidase by R. equi and sphingomyelinase D by A. haemolyticum) that are toxic to animals and humans and damaging to mammalian cell membranes. The participation of the cytotoxins in pathogenicity is discussed. Greater awareness of the properties of these two bacteria may promote faster, more accurate diagnoses and better clinical management.