Assessing the Economic Value of Clinical Artificial Intelligence: Challenges and Opportunities

OBJECTIVES

Clinical artificial intelligence (AI) is a novel technology, and few economic evaluations have focused on it to date. Before its wider implementation, it is important to highlight the aspects of AI that challenge traditional health technology assessment methods.

METHODS

We used an existing broad value framework to assess potential ways AI can provide good value for money. We also developed a rubric of how economic evaluations of AI should vary depending on the case of its use.

RESULTS

We found that the measurement of core elements of value-health outcomes and cost-are complicated by AI because its generalizability across different populations is often unclear and because its use may necessitate reconfigured clinical processes. Clinicians' productivity may improve when AI is used. If poorly implemented though, AI may also cause clinicians' workload to increase. Some AI has been found to exacerbate health disparities. Nevertheless, AI may promote equity by expanding access to medical care and, when properly trained, providing unbiased diagnoses and prognoses. The approach to assessment of AI should vary based on its use case: AI that creates new clinical possibilities can improve outcomes, but regulation and evidence collection may be difficult; AI that extends clinical expertise can reduce disparities and lower costs but may result in overuse; and AI that automates clinicians' work can improve productivity but may reduce skills.

CONCLUSIONS

The potential uses of clinical AI create challenges for health technology assessment methods originally developed for pharmaceuticals and medical devices. Health economists should be prepared to examine data collection and methods used to train AI, as these may impact its future value.

Silicon Photoelectrode Thermodynamics and Hydrogen Evolution Kinetics Measured by Intensity-Modulated High-Frequency Resistivity Impedance Spectroscopy

We present an impedance technique based on light intensity-modulated high-frequency resistivity (IMHFR) that provides a new way to elucidate both the thermodynamics and kinetics in complex semiconductor photoelectrodes. We apply IMHFR to probe electrode interfacial energetics on oxide-modified semiconductor surfaces frequently used to improve the stability and efficiency of photoelectrochemical water splitting systems. Combined with current density-voltage measurements, the technique quantifies the overpotential for proton reduction relative to its thermodynamic potential in Si photocathodes coated with three oxides (SiO_x, TiO₂, and Al₂O₃) and a Pt catalyst. In pH 7 electrolyte, the flatband potentials of TiO₂- and Al₂O₃-coated Si electrodes are negative relative to samples with native SiO_x, indicating that SiO_x is a better protective layer against oxidative electrochemical corrosion than ALD-deposited crystalline TiO₂ or Al₂O₃. Adding a Pt catalyst to SiO_x/Si minimizes proton reduction overpotential losses but at the expense of a reduction in available energy characterized by a more negative flatband potential relative to catalyst-free SiO_x/Si.

Synthesis and Spectroscopy of Silver-Doped PbSe Quantum Dots

Electronic impurity doping of bulk semiconductors is an essential component of semiconductor science and technology. Yet there are only a handful of studies demonstrating control of electronic impurities in semiconductor nanocrystals. Here, we studied electronic impurity doping of colloidal PbSe quantum dots (QDs) using a postsynthetic cation exchange reaction in which Pb is exchanged for Ag. We found that varying the concentration of dopants exposed to the as-synthesized PbSe QDs controls the extent of exchange. The electronic impurity doped QDs exhibit the fundamental spectroscopic signatures associated with injecting a free charge carrier into a QD under equilibrium conditions, including a bleach of the first exciton transition and the appearance of a quantum-confined, low-energy intraband absorption feature. Photoelectron spectroscopy confirms that Ag acts as a p-type dopant for PbSe QDs and infrared spectroscopy is consistent with k·p calculations of the size-dependent intraband transition energy. We find that to bleach the first exciton transition by an average of 1 carrier per QD requires that approximately 10% of the Pb be replaced by Ag. We hypothesize that the majority of incorporated Ag remains at the QD surface and does not interact with the core electronic states of the QD. Instead, the excess Ag at the surface promotes the incorporation of <1% Ag into the QD core where it causes p-type doping behavior.

Covalent Surface Modification of Gallium Arsenide Photocathodes for Water Splitting in Highly Acidic Electrolyte

Efficient water splitting using light as the only energy input requires stable semiconductor electrodes with favorable energetics for the water-oxidation and proton-reduction reactions. Strategies to tune electrode potentials using molecular dipoles adsorbed to the semiconductor surface have been pursued for decades but are often based on weak interactions and quickly react to desorb the molecule under conditions relevant to sustained photoelectrolysis. Here, we show that covalent attachment of fluorinated, aromatic molecules to p-GaAs(1 0 0) surfaces can be employed to tune the photocurrent onset potentials of p-GaAs(1 0 0) photocathodes and reduce the external energy required for water splitting. Results indicate that initial photocurrent onset potentials can be shifted by nearly 150 mV in pH -0.5 electrolyte under 1 Sun (1000 W m^-2 ) illumination resulting from the covalently bound surface dipole. Though X-ray photoelectron spectroscopy analysis reveals that the covalent molecular dipole attachment is not robust under extended 50 h photoelectrolysis, the modified surface delays arsenic oxide formation that results in a p-GaAs(1 0 0) photoelectrode operating at a sustained photocurrent density of -20.5 mA cm^-2 within -0.5 V of the reversible hydrogen electrode.

Proton Reduction Using a Hydrogenase-Modified Nanoporous Black Silicon Photoelectrode

Metalloenzymes featuring earth-abundant metal-based cores exhibit rates for catalytic processes such as hydrogen evolution comparable to those of noble metals. Realizing these superb catalytic properties in artificial systems is challenging owing to the difficulty of effectively interfacing metalloenzymes with an electrode surface in a manner that supports efficient charge-transfer. Here, we demonstrate that a nanoporous "black" silicon (b-Si) photocathode provides a unique interface for binding an adsorbed [FeFe]-hydrogenase enzyme ([FeFe]-H2ase). The resulting [FeFe]-H2ase/b-Si photoelectrode displays a 280 mV more positive onset potential for hydrogen generation than bare b-Si without hydrogenase, similar to that observed for a b-Si/Pt photoelectrode at the same light intensity. Additionally, we show that this H2ase/b-Si electrode exhibits a turnover frequency of ≥1300 s(-1) and a turnover number above 10(7) and sustains current densities of at least 1 mA/cm(2) based on the actual surface area of the electrode (not the smaller projected geometric area), orders of magnitude greater than that observed for previous enzyme-catalyzed electrodes. While the long-term stability of hydrogenase on the b-Si surface remains too low for practical applications, this work extends the proof-of-concept that biologically derived metalloenzymes can be interfaced with inorganic substrates to support technologically relevant current densities.

All-Inorganic Germanium Nanocrystal Films by Cationic Ligand Exchange

We introduce a new paradigm for group IV nanocrystal surface chemistry based on room temperature surface activation that enables ionic ligand exchange. Germanium nanocrystals synthesized in a gas-phase plasma reactor are functionalized with labile, cationic alkylammonium ligands rather than with traditional covalently bound groups. We employ Fourier transform infrared and (1)H nuclear magnetic resonance spectroscopies to demonstrate the alkylammonium ligands are freely exchanged on the germanium nanocrystal surface with a variety of cationic ligands, including short inorganic ligands such as ammonium and alkali metal cations. This ionic ligand exchange chemistry is used to demonstrate enhanced transport in germanium nanocrystal films following ligand exchange as well as the first photovoltaic device based on an all-inorganic germanium nanocrystal absorber layer cast from solution. This new ligand chemistry should accelerate progress in utilizing germanium and other group IV nanocrystals for optoelectronic applications.

Oxidatively stable nanoporous silicon photocathodes with enhanced onset voltage for photoelectrochemical proton reduction

Stable and high-performance nanoporous "black silicon" photoelectrodes with electrolessly deposited Pt nanoparticle (NP) catalysts are made with two metal-assisted etching steps. Doubly etched samples exhibit an ∼300 mV positive shift in photocurrent onset for photoelectrochemical proton reduction compared to oxide-free planar Si with identical catalysts. We find that the photocurrent onset voltage of black Si photocathodes prepared from single-crystal planar Si wafers by an Ag-assisted etching process increases in oxidative environments (e.g., aqueous electrolyte) owing to a positive flat-band potential shift caused by surface oxidation. However, within 24 h, the surface oxide layer becomes a kinetic barrier to interfacial charge transfer that inhibits proton reduction. To mitigate this issue, we developed a novel second Pt-assisted etch process that buries the Pt NPs deep into the nanoporous Si surface. This second etch shifts the onset voltage positively, from +0.25 V to +0.4 V versus reversible hydrogen electrode, and reduces the charge-transfer resistance with no performance decrease seen for at least two months. PEC performance was stable owing to Pt NP catalysts that were buried deeply in the photoelectrode by the second etch, below a thick surface layer comprised primarily of amorphous SiO2 along with some degree of remaining crystalline Si as observed by scanning and transmission electron micrographs. Electrochemical impedance studies reveal that the second etch leads to a considerably smaller interfacial charge-transfer resistance than samples without the additional etch, suggesting that burying the Pt NPs improves the interfacial contact to the crystalline silicon surface.

Electrical transport and grain growth in solution-cast, chloride-terminated cadmium selenide nanocrystal thin films

We report the evolution of electrical transport and grain size during the sintering of thin films spin-cast from soluble phosphine and amine-bound, chloride-terminated cadmium selenide nanocrystals. Sintering of the nanocrystals occurs in three distinct stages as the annealing temperature is increased: (1) reversible desorption of the organic ligands (≤150 °C), (2) irreversible particle fusion (200-300 °C), and (3) ripening of the grains to >5 nm domains (>200 °C). Grain growth occurs at 200 °C in films with 8 atom % Cl(-), while films with 3 atom % Cl(-) resist growth until 300 °C. Fused nanocrystalline thin films (grain size = 4.5-5.5 nm) on thermally grown silicon dioxide gate dielectrics produce field-effect transistors with electron mobilities as high as 25 cm(2)/(Vs) and on/off ratios of 10(5) with less than 0.5 V hysteresis in threshold voltage without the addition of indium.

Ligand exchange and the stoichiometry of metal chalcogenide nanocrystals: spectroscopic observation of facile metal-carboxylate displacement and binding

We demonstrate that metal carboxylate complexes (L-M(O2CR)2, R = oleyl, tetradecyl, M = Cd, Pb) are readily displaced from carboxylate-terminated ME nanocrystals (ME = CdSe, CdS, PbSe, PbS) by various Lewis bases (L = tri-n-butylamine, tetrahydrofuran, tetradecanol, N,N-dimethyl-n-butylamine, tri-n-butylphosphine, N,N,N',N'-tetramethylbutylene-1,4-diamine, pyridine, N,N,N',N'-tetramethylethylene-1,2-diamine, n-octylamine). The relative displacement potency is measured by (1)H NMR spectroscopy and depends most strongly on geometric factors such as sterics and chelation, although also on the hard/soft match with the cadmium ion. The results suggest that ligands displace L-M(O2CR)2 by cooperatively complexing the displaced metal ion as well as the nanocrystal. Removal of up to 90% of surface-bound Cd(O2CR)2 from CdSe and CdS nanocrystals decreases the Cd/Se ratio from 1.1 ± 0.06 to 1.0 ± 0.05, broadens the 1S(e)-2S(3/2h) absorption, and decreases the photoluminescence quantum yield (PLQY) from 10% to <1% (CdSe) and from 20% to <1% (CdS). These changes are partially reversed upon rebinding of M(O2CR)2 at room temperature (∼60%) and fully reversed at elevated temperature. A model is proposed in which electron-accepting M(O2CR)2 complexes (Z-type ligands) reversibly bind to nanocrystals, leading to a range of stoichiometries for a given core size. The results demonstrate that nanocrystals lack a single chemical formula, but are instead dynamic structures with concentration-dependent compositions. The importance of these findings to the synthesis and purification of nanocrystals as well as ligand exchange reactions is discussed.

Broadband absorbing bulk heterojunction photovoltaics using low-bandgap solution-processed quantum dots

We describe bulk heterojunction (BHJ) solar cells containing blends of colloidal PbS nanocrystal quantum dots with several new donor-acceptor conjugated polymers. Using photoinduced absorption spectroscopy we found that blends of PbS quantum dots with one polymer, poly(2,3-didecyl-quinoxaline-5,8-diyl-alt-N-octyldithieno[3,2-b:2',3'-d]pyrrole) (PDTPQx), produce significantly more photoinduced charge than blends of PbS with the other donor-acceptor polymers or with traditionally studied polymers like [2-methoxy-5-(3',7'-dimethyloctyloxy)-para-phenylene vinylene] (MDMO-PPV) and poly(3-hexylthiophene) (P3HT). Photovoltaic devices made with PDTPQx/PbS blends exhibit power conversion efficiencies 10-100 times larger than previously reported BHJ blends made with IR-absorbing quantum dots.

Absence of photoinduced charge transfer in blends of PbSe quantum dots and conjugated polymers

We use photoluminescence (PL) quenching and photoinduced absorption (PIA) spectroscopy to study charge transfer in bulk heterojunction blends of PbSe quantum dots with the semiconducting polymers poly-3-hexylthiophene (P3HT) and poly[2-methoxy-5-(3',7'-dimethyloctyloxy)-para-phenylene vinylene] (MDMO-PPV). PIA spectra from the PbSe blends are compared to spectra from similar blends of the polymers with phenyl-C(61)-butyric acid methyl ester (PCBM) and blends with CdSe quantum dots. We find that the MDMO-PPV PL is quenched, and the PL lifetime is shortened upon addition of PbSe quantum dots, while the PL of the P3HT is unaffected upon blending. However, for PbSe blends with both polymers, the PIA spectra show very little polaronic signal, suggesting that few, if any, long-lived charges are being produced by photoinduced charge transfer.

Passive membrane properties and inward calcium current of human uterine smooth muscle cells

Human myometrium was obtained at the time of cesarean delivery by means of excisional biopsy from the upper margin of the uterine incision through the lower uterine segment. Isolated smooth muscle cells were obtained by collagenase treatment. With the whole-cell patch clamp technique, voltage-clamp and current-clamp experiments were performed. Regenerative action potentials were seen in 3 mmol/L calcium bathing solution. Voltage-clamp experiments demonstrated the presence of transient inward currents and large outward currents. Inward currents were isolated with the use of tetraethylammonium ion as a blocking agent of outward current. In 3 mmol/L calcium inward current began at -60 mV and maximum inward current occurred at -40 mV. Maximum inward current density during the rising phase of the action potential was 1.0 microA/cm2 in 3 mmol/L calcium and was unaffected by reduction of sodium concentration in the bathing solution.

Morphologic and electrophysiologic characterization of isolated pregnant human myometrial cells

Myometrium was obtained from pregnant volunteers by biopsy of the upper margin of the uterine incision at the time of cesarean section. A multistep enzymatic digestion with collagenase, trypsin, protease, and deoxyribonuclease yielded viable cells capable of contraction. Primary monolayer culture was carried out in the presence of human pregnant serum. Electron microscopic examination of freshly isolated and cultured cells revealed an ultrastructure indicative of smooth muscle cells. Intracellular microelectrode studies were performed on freshly isolated cells. Passive membrane properties were: resting membrane potential, -49.4 mV; specific membrane resistance, 6.06 kohms-cm2; specific membrane capacitance, 1.57 microfarad per square centimeter. Outward-going rectification was observed in response to depolarizing current pulses. Regenerative action potentials were not observed; however, transient voltage responses were elicited after depolarizing, but not hyperpolarizing, current pulses. These studies characterize a human tissue preparation that is applicable to electrophysiologic investigation of the control of uterine function.

The effects of aminophylline and nifedipine on contractility of isolated pregnant human myometrium

To characterize the effects of aminophylline and nifedipine on pregnant human myometrium, in vitro contractility studies were performed on myometrial strips obtained at cesarean section. The strips were stimulated with oxytocin (800 mU/L) to simulate labor and then were exposed to increasing concentrations of aminophylline (40, 100, and 400 mumol/L) or nifedipine (5, 10, and 20 micrograms/L). Both drugs produced a dose-related decrease in contraction strength, as measured by the time-integrated force of contraction. Aminophylline lengthened the period of contraction in a manner that was not dose dependent. Low-dose nifedipine (5 micrograms/L) increased the period of contraction, but higher doses had no effect on frequency. Both drugs produced a net reduction in the effectiveness of labor, as measured by the average force (time-integrated force divided by period). These results indicate that both aminophylline and nifedipine may be clinically useful tocolytic agents.

Ultrastructure and morphometry of the stomach muscle of Amphiuma tridactylum

An ultrastructural and stereological examination was performed on stomach smooth muscle of the salamander Amphiuma. This tissue has very large cells, ranging up to 12 X 1500 micron when relaxed. The extracellular space is 31% of the tissue volume, and the tissue contains 84.6% water. These values are similar to those of other amphibian and mammalian gastrointestinal smooth muscle. The cells possess the usual smooth muscle organelles. Thick, thin and intermediate filaments are present, along with membrane-associated and cytoplasmic dense regions. There is a well-developed sarcoplasmic reticulum and many microtubules. Caveolae are found in rows along the cellular surface; the caveolae increase the cellular surface area by about 70%. The ratio mean volume: surface area of the cells is 1.26 micron. This tissue appears to be typical of gastrointestinal smooth muscle, with the exception of the very large size of the cells.

Hyperpolarization in mouse parathyroid cells by low calcium

Intracellular microelectrodes were used to record the resting membrane potential of mouse parathyroid cells in vitro. The mean value of the membrane potential in 2.5 mM calcium was -20 mV. Exposure to low-calcium solutions (1.5 mM) caused rapid hyperpolarization to a mean value of -50 mV. The relationship between extracellular calcium and the membrane potential was sigmoid, and the transition between high and low intracellular potentials occurred between 1.5 mM and 2.25 mM calcium. Magnesium, manganese, and lanthanum reversed the low-calcium hyperpolarization. In 1.5 mM calcium solutions, the relationship between external potassium (greater than 10 mM) and the membrane potential was 46 mV per 10-fold change in extracellular potassium. In 2.5 mM calcium solutions, the resting membrane potential was independent of the external potassium concentration. It is concluded that the resting membrane potential of mouse parathyroid cells is highly sensitive to the extracellular concentration of calcium and calcium-like ions. With the low-calcium secretory stimulus, hyperpolarization is accompanied by the development of strong dependence of the resting membrane potential on extracellular potassium levels.

Intercellular communication in the rat anterior pituitary gland. An in vivo and in vitro study

The concept of "stimulus-secretion coupling" suggested by Douglas and co-workers to explain the events related to monamine discharge by the adrenal medulla (5, 7) may be applied to other endocrine tissues, such as adrenal cortex (36), pancreatic islets (4), and magnocellular hypothalamic neurons (6), which exhibit a similar ion-dependent process of hormone elaboration. In addition, they share another feature, that of joining neighbor cells via membrane junctions (12, 26, and Fletcher, unpublished observation). Given this, and the reports that hormone secretion by the pars distalis also involves a secretagogue-induced decrease in membrane bioelectric potential accompanied by a rise in cellular [Ca++] (27, 34, 41), it was appropriate to test the possibility that cells of the anterior pituitary gland are united by junctions.

Studies on calcium and sodium in uterine smooth muscle excitation under current-clamp and voltage-clamp conditions

The objective of these studies was to define the roles of calcium and sodium in uterine smooth muscle excitation. The double sucrose-gap technique was used for current-clamp and voltage-clamp experiments. It was shown that neither sodium nor calcium alone is capable of supporting excitation in estrogen-dominated uterine smooth muscle. Calcium dependence was explained in part by increased membrane "leakage" current in calcium-free solution and calcium control of the voltage dependence of the early transient conductance. High concentrations of TTX did not affect the magnitude of the peak transient current while La(+++), Mn(++), and Co(++) greatly reduced or abolished it and decreased the steady-state current. From these and other data it was concluded that the regenerative mechanism in uterine smooth muscle has the functional characteristics of a single transient conductance channel whose activation requires the presence of both sodium and calcium. Insensitivity to TTX indicates that the molecular structure of the channel is unlike that in certain sodium-dependent systems, while the effects of La(+++), Mn(++), Co(++), and Ca(++) reveal a similar dependence of conductances on extracellular polyvalent cations.

Voltage-clamp studies on uterine smooth muscle

These studies have developed and tested an experimental approach to the study of membrane ionic conductance mechanisms in strips of uterine smooth muscle. The experimental and theoretical basis for applying the double sucrose-gap technique is described along with the limitations of this system. Nonpropagating membrane action potentials were produced in response to depolarizing current pulses under current-clamp conditions. The stepwise change of membrane potential under voltage-clamp conditions resulted in a family of ionic currents with voltage- and time-dependent characteristics. In sodium-free solution the peak transient current decreased and its equilibrium potential shifted along the voltage axis toward a more negative internal potential. These studies indicate a sodium-dependent, regenerative excitation mechanism.

Comparison of tetrodotoxin and procaine in internally perfused squid giant axons

Squid giant axons were internally perfused with tetrodotoxin and procaine, and excitability and electrical properties were studied by means of current-clamp and sucrose-gap voltage-clamp methods. Internally perfused tetrodotoxin was virtually without effect on the resting potential, the action potential, the early transient membrane ionic current, and the late steady-state membrane ionic current even at very high concentrations (1,000-10,000 nM) for a long period of time (up to 36 min). Externally applied tetrodotoxin at a concentration of 100 nM blocked the action potential and the early transient current in 2-3 min. Internally perfused procaine at concentrations of 1-10 mM reversibly depressed or blocked the action potential with an accompanying hyperpolarization of 2-4 mv, and inhibited both the early transient and late steady-state currents to the same extent. The time to peak early transient current was increased. The present results and the insolubility of tetrodotoxin in lipids have led to the conclusion that the gate controlling the flow of sodium ions through channels is located on the outer surface of the nerve membrane.

Basis of tetrodotoxin's selectivity in blockage of squid axons

The blockage of nerve activity by tetrodotoxin is unusually potent and specific. Our experiments were designed to distinguish whether its specificity of action was based on the identification of ions, the direction of cation flow, or differences in the early transient and late steady conductance pathways. Alkali cations were substituted for sodium in the sea water, bathing an "artificial node" in a voltage-clamped squid axon. When tetrodotoxin was added to the artificial sea waters at a concentration of 100 to 150 mM, it was found to always block the flow of cations through the early transient channel, both inward and outward, but it never blocked the flow of ions using the late steady pathway. We conclude that the selectivity of tetrodotoxin is based on some difference in these two channels.

Tetrodotoxin does not block excitation from inside the nerve membrane

Tetrodotoxin does not block the action potential or membrane sodium current when internally perfused through the giant axon of a squid at much higher concentrations than those required for blocking by external application. It is suggested that the gate for the sodium channel is located on the exterior surface of the axon, because tetrodotoxin is not lipid soluble.