TL;DR: Think PASCAL for critical and wildly complex life-support systems; operating in the Internet age.
Having spent the last 10 years working in healthcare, I cannot overemphasize the ever growing cataclysm-to-be built upon MUMPS. One might then associate MUMPS the language with Mumps the disease (sans vaccine).
Worldwide, most major Electronic Health Records (EHR) systems are fundamentally based on MUMPS. Since the language largely predates all modern programming and system design principles like logic-data abstraction, imagine the mess created when various clinical applications (i.e. apps for pediatrics vs. oncology vs. pharmacy vs. radiology vs. geriatrics vs. etc.) are independently created and loosely cobbled together into a complex system - duplicative data representation and non-deterministic 'macro' processes/procedures are woefully common.
As cited in the Wikipedia MUMPS page, MUMPS is foundational to the largest EHRs (ref. Top 10 [US] EHR vendors by overall market share. In my opinion, this isn't so much because MUMPS is superior in anyway but because there really is 40+ years of codebase out there; much of it still in use today. Consider the US Indian Health Service (IHS) as example. For FY2014, IHS spent a total of $98M for IT. Although recognizing its own RPMS is hopelessly flawed, the agency estimates a capital investment >$150M and 15+ years to transition to newer technology. All the while, they must maintain and operate the current RPMS for ongoing healthcare delivery. Suffice to say and for the foreseeable future, IHS is hopelessly bound to the incumbent RPMS.
Meanwhile, "extensions" to MUMPS continue to proliferate. Since MUMPS heralds from the days of text-only dumb terminals, there is no in-language accommodation for graphical user interfaces and the common controls associated with event driven systems. This has led to many language extensions created for presentation 'wrapping' the underlying MUMPS output. And since MUMPS precedes modern design principles, newly minted programmers, admins and integrators must learn everything "from scratch" and, most often, in defiance/conflict of tenets they have been taught. Tenets and principles developed in response to hard lessons learned over the last 40 years of widespread IT experiences. These issues are exacerbated because those with the greatest experience are largely retired (or soon to be) and there are no significant opportunities for educating younger professionals except through other than "real life" - using live/production systems and actual persons' medical information.