Artificial Intelligence and Black‐Box Medical Decisions: Accuracy versus Explainability

Although decision‐making algorithms are not new to medicine, the availability of vast stores of medical data, gains in computing power, and breakthroughs in machine learning are accelerating the pace of their development, expanding the range of questions they can address, and increasing their predictive power. In many cases, however, the most powerful machine learning techniques purchase diagnostic or predictive accuracy at the expense of our ability to access “the knowledge within the machine.” Without an explanation in terms of reasons or a rationale for particular decisions in individual cases, some commentators regard ceding medical decision‐making to black box systems as contravening the profound moral responsibilities of clinicians. I argue, however, that opaque decisions are more common in medicine than critics realize. Moreover, as Aristotle noted over two millennia ago, when our knowledge of causal systems is incomplete and precarious—as it often is in medicine—the ability to explain how results are produced can be less important than the ability to produce such results and empirically verify their accuracy.     

A pulse reflectometer for routine monitoring of transmitted and reflected power in physiological NMR studies

A simple circuit design is described which provides LED readout of transmitted and reflected pulsed rf power. The utility of monitoring these variables in physiological NMR studies is discussed.     

Colonoscopy and bacteraemia: an experience in 50 patients

There is little consensus concerning the incidence of bacteraemia during colonoscopy and the need for antibiotic prophylaxis in susceptible patients. Hepatic abscesses in one patient which may have been related to prior colonoscopic examinations led the authors to carry out a prospective study of 50 patients undergoing colonoscopy. Multiple blood cultures were carried out to maximise the positive yield of transient bacteraemia and to attempt to determine the time when bacteraemia is most likely to occur. Five patients had positive blood cultures. In two patients S epidermidis was isolated, but only from the precolonoscopic blood sample. In three subjects enteric organisms were cultured from blood samples obtained during the procedure. In one of these three the same organism was cultured from the preendoscopic blood sample so that in only two patients (4%) could the bacteraemia be attributed to the colonoscopy. These results would suggest that the risk of bacteraemia during colonoscopy is low.     

Is infrainguinal bypass grafting successful following failed angioplasty?

OBJECTIVES: Angioplasty is often used in the management of lower limb ischaemia and can reduce the need for infrainguinal bypass in some patients. There is an associated failure rate with this technique and bypass surgery is often used in this situation as a secondary limb salvage procedure. We aimed to evaluate the outcome of infrainguinal bypass grafting following failed attempt at angioplasty. METHODS: All cases of infrainguinal bypass at a single centre over a seven year period were identified and notes reviewed. Cases were divided into four groups according to their indication for surgery; acute ischaemia, chronic critical ischaemia, failed angioplasty and an 'other' group including aneurysmal disease and claudicants. The failed angioplasty group was compared with the other three groups. Survival analysis was performed using Kaplan Meier curves and groups compared in terms of long term patency and survival. RESULTS: Primary patency was 61.2% in the failed angioplasty group at 12 months compared with 60.6% in the other groups (P=1.11). There was also no significant difference in primary patency at 60 months (50% vs 40.6%, P=0.26). Survival at 12 months was also comparable between the groups (failed angioplasty group 74.2% compared with 77.3% in the other groups, P=0.662) as was 60 months survival (33.3% and 35.4% respectively, P=0.166). DISCUSSION: In this study, outcome of infrainguinal bypass following failed angioplasty was comparable to outcome of surgery performed for another indication. This paper supports the use of distal bypass surgery for limb salvage in cases where minimal access techniques have failed.     

Suppurative corneal ulceration in Bangladesh

Suppurative keratitis is an important preventable cause of blindness, particularly in the developing world. This study analyses 142 cases of suppurative keratitis referred to Chittagong Eye Infirmary, Bangladesh. Some 53.5% of cases were bacterial and 35.9% were fungal. The five most common pathogens were: Pseudomonas sp. 24%, Streptococcus pneumoniae 17%, Aspergillus sp. 13%, Fusarium sp. 7% and Curvularia sp. 6%. Gram stain and culture results were consistent in 62.6% of cases. Previous antibiotic treatment was a significant factor for failure of culture isolation and less so for Gram stain failure. On Gram stain, 55.9% of pseudomonal cases were missed, but only 2% of fungal cases were missed. Over all, Gram stain had a sensitivity of 62% and positive predictive value of 84% for bacterial cases, and 98% and 94% for fungal cases, respectively. Fungal ulcers were typically filamentous, but an antecedent history of trauma was not common. The most frequent injury was due to rice grains, but the inoculum appeared to be introduced during eye washing with contaminated water. Pseudomonal ulcers occurred most frequently in the monsoon season, and Fusarium cases were seen only in the hot, dry season.     

Prospective study of relative weight, height, and risk of breast cancer

We examined relative weight and height in relation to subsequent breast cancer risk among 115,534 women 30 to 55 years of age and free from cancer in 1976. By 1984, six hundred fifty-eight premenopausal and 420 postmenopausal breast cancers were documented during 734,716 person-years. Among premenopausal women, risk of breast cancer decreased significantly with increasing relative weight (relative risk for the highest category was 0.6). A similar inverse association was seen for recalled relative weight at 18 years of age. Postmenopausal breast cancer was not associated with relative weight, either recent or at age 18. Height was not associated with breast cancer risk among premenopausal women and only weakly related among postmenopausal women. These data suggest that obesity among premenopausal and early postmenopausal women does not increase breast cancer risk substantially.